This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
) « Ipip/ipo where Ip is the circulating current in one of the stable minima of ip, we end up with the classical equation of motion of the qubit including the backaction and the friction induced from the SQUID -CeS ( f ^ ) due to the diminished radius of the trajectory starting from the initial state uz = +1. The impact of the relaxation/excitation generating part is furthermore depending on cos ij> as well as sin \ , the azimuth angle of the qubits present position.
w2 + 2TT 2 M 2 Q J 2 i a s tan 2
= -d
7in,0
' itp0u$o
V
(20)
We encode this form using the Fourier transform of the differential operators as D(uj)ip{w) = —dU/d(f. We can use the prescription given in 155 and identify the spectral function for the continuous, classical model as JCont = ImD(u>). From there, we can do the two-state approximation for the particle in a double well 140 and find J(u) = JCont in analogy to 156
76
This equation has a remarkable consequence: It predicts a well-defined "off'-state at /bias = 0, i.e. as for any good detector, the backaction can be switched off in order to prevent the Zeno effect. In practice, one has to take into account the double-layer structure, so either the preferred working point is at finite current, or the double layer is shorted by a fourth junction 90 , see also section 5.3. Note that, even though we have assumed a shunt only across the whole device, this method can be extended to describe shunts at the individual SQUID junctions as well 157 . 7.4. Qubit Dynamics
under the Influence
of
Decoherence
Prom J(w), we can analyze the dynamics of the system by studying the reduced density matrix, i.e. the density matrix of the full system where the details of the environment have been integrated out, by a number of different methods. The low damping limit, J(w)/w
77
7.5. Application:
Engineering
the Measurement
Apparatus
From Eq. (21) we see that engineering the decoherence induced by the measurement apparatus essentially means engineering Zeg. We specifically focus on the contributions due to the measurement apparatus. In this section, we are going to outline and compare several options suggested in literature. We assume a perfect current source that ramps the bias current -fbias through the SQUID. The fact that the current source is non-ideal, and that the wiring to the SQUID chip has an impedance is all modeled by the impedance Z(u>). The wiring can be engineered such that for a very wide frequency range the impedance Z(u>) is on the order of the vacuum impedance, and can be modeled by its real part Ri. It typically has a value of 100 fi. It has been suggested 30 to overdamp the SQUID by making the shunt circuit a simple resistor Z(UJ) = R$ with i?s -C ^/L^n/2Cj. This is inspired 163 by an analogous setup for charge qubits, . Following the parameters given in 156 , we find aR = 0.08ft and UJLR/R = 8.3GHz/fi. Next, we consider a large superconducting capacitive shunt (Fig. 15a, as implemented in Refs. 7 5 ' 1 6 4 ). The C shunt only makes the effective mass of the SQUID's external phase 7 e x very heavy. The total impedance Zes(uj) and J{OJ) are modeled as before, see Fig. 16. As limiting values, we find
f <£&, f o r W < W L C Re{Zeff(u;)} « I
Ru
for u = UJLC
(22)
I I ^ c b v for u, » w L C We can observe that this circuit is a weakly damped LC-oscillator and it is clear from (12) and (21) that one should keep its resonance frequency WLC = l/V-^jCsh; where Re{Zeff(u;)} has a maximum, away from the qubit's resonance w res = v/ft- This is usually done by chosing WLC ^ wresFor a C-shunted circuit with LULC <S w r e s , this yields for J(OJ « WLC) T/
,
(2TT)2 fMIp\2
r2
2 / ^
1
One has to be aware of the fact that at these high switching current values the linearization of the junction as a kinetic inductor may underestimate the actual noise. In that regime, phase diffusion between different minima of the washboard potential also becomes relevant and changes the noise properties 165>166. As a complementary perspective, the C-shunted SQUID can be seen as a, generally nonlinear, cavity with small leakage rate 114 - 162 .
78
a
A «
dv
bias
V
Fig. 15. Circuit models for the C-shunted DC-SQUID (a) and the flC-shunted DCSQUID (b). The SQUID is modeled as an inductance Lj. A shunt circuit, the superconducting capacitor C s h or the i? s h-Csh series, is fabricated on chip very close to the SQUID. The noise that couples to the qubit results from Johnson-Nyquist voltage noise 5V from the circuit's total impedance Zeg. Zeg is formed by a parallel combination of the impedances of the leads Z{, the shunt and the SQUID, such that Z~g = l/Zi + l/(Rsh + l/iwCsh) + 1/iuiLj, with Rsh = 0 for the circuit (a)
As an alternative we will consider a shunt that is a series combination of a capacitor and a resistor (Fig. 15b) (i?C-shunted SQUID). The RC shunt also adds damping at the plasma frequency of the SQUID, which is needed for realizing a high resolution of the SQUID readout (i. e. for narrow switching-current histograms) 165 . The total impedance Zt(oj) of the two measurement circuits are modeled as in Fig. 15. For the circuit with the RC shunt Ri
Re{Zt(uj)}
for LJ -C OJLC
(24)
79
C shunt
R C shunt 10'
b
10° •
10 2 m4-
10°
10
(0/27t (Hz)
1010
107
10
10
10
(0/271 (Hz)
Fig. 16. A typical Re{Z t (u>)} for the C-shunted SQUID (a) and the .RC-shunted SQUID (b), and corresponding J{oS) in (c) and (d) respectively. For comparison, the dashed line in (c) shows a simple Ohmic spectrum, 7(a>) = aui with exponential cut off uic/27r = 0.5 GHz and a = 0.00062. The parameters used here are Ip — 500 nA and T = 30 mK. The SQUID with 2Ico = 200 nA is operated at / = 0.75 w and current biased at 120 nA, a typical value for switching of the C-shunted circuit (the .RC-shunted circuit switches at higher current values). The mutual inductance M = 8 pH (i. e. MIp/$o = 0.002). The shunt is C s h = 30 pF and for the RC shunt Rsil = 10 H. The leads are modeled by Ri = 100 Q
The difference mainly concerns frequencies w > LJLC where the C-shunted circuit has a stronger cutoff in Re{Zeff(ui)}, and thereby a relaxation rate, that is several orders lower than for the .RC-shunted circuit. Given the values of J(w) from Fig. 16, one can directly see that an .RC-shunted circuit with otherwise similar parameters yields at wres/2-K — 10 GHz relaxation times that are about four orders of magnitude shorter. Let us remark that a different parameter regime of the RC shunt has been studied in Ref. 157 . Here, the RC shunts are higher in resistance and allow to enter the quantum tunneling regime for the SQUID. Also, they are
80
not applied across the device, but to the individual junctions. The steering of the minimum width and the histograms has been studied in great detail and good agreement between theory and experiment has been obtained. However, it turned out that the best device performance will be achieved for R = 0, a pure C-shunt.
7.6. Discrete
noise
Not all environments can be described this way. A prominent counterexample are localized modes with a bounded spectrum such as spins 167 and structural fluctuations generating classical telegraph noise 1 6 8 _ 1 7 1 ) which are inherently non-Gaussian. One expects from the central limit theorem, that large ensembles of such fluctuators usually behave Gaussian again 172,130 , but details of this transition are only partially understood at present 169,173
Most research has concentrated on decoupling devices from external noise sources such as electromagnetic noise generated by control and measurement apparatus 156 . On the other hand, there inevitably are internal noise sources because the fabrication of gates, tunnel junctions, and other functional components creates defects in the underlying crystal. Prominent examples of such defects are background charges in charge-based devices or cricital current fluctuations in flux-based devices 174 - 175 . A clear signature of such defects is telegraph noise in the case of a few defects or 1//-noise in the case of a larger ensemble 176 . With the growing success in engineering the electromagnetic environment, these defects are becoming more and more the key limiting sources of decoherence. Such defects do not fall in the large class of noise sources that can be approximated well as a bosonic bath, and this fact complicates their analysis. Localized noise sources with bounded spectra like the defects in which we are interested produce noise that is significantly non-Gaussian. Theories treating large ensembles of non-Gaussian noise sources have been presented 168,167 f j o w e v e r ; with the ongoing improvement in nanofabrication technology, it is realistic to consider the case where non-Gaussian noise sources are reduced down to only a single one or a few per device. This is the case we treat here, and thus the defects find a more realistic representation as a small set of bistable fluctuators (henceforth abbreviated bfls). In principle, this approach can be extended to larger sets of bfls with a range of different mean switching times (e.g., an ensemble with an exponential distribution of switching times that produces 1//-noise 99>173>171).
81
7.7. Model of the bistable limit
fluctuator
in its
semiclassical
We describe the bfl-noise influenced evolution of the qubit in its semiclassical limit by using a stochastic Schrodinger equation 177>178 with the timedependent effective Hamiltonian Hf(t)
= Hq + Hnoise(t)
(25)
tfq = heq&1 + HAqa* Hnoise(t)
= haal
(26)
£bfl(*)
(27)
where eq and A q define the free (noiseless) qubit dynamics. £bfl(£) denotes a function randomly switching between ±1 (see Fig. 17), which represents a telegraph noise signal. The switching events follow a symmetrical Poisson process, i.e., the probabilities of the bfl switching from + 1 to —1 or —1 to +1 are the same and equal in time. The Poisson process is characterized by the mean time separation Tbfl between two bfl flips. The coupling amplitude to the qubit in frequency units is a. Starting with an arbitrary initial state of the qubit, represented by some given point on the Bloch sphere, we can numerically integrate the corresponding stochastic differential equation and obtain the corresponding random walk on the Bloch sphere a{t) = Texp l-i/h
J Hf{s)ds)
ff(0)
(28)
with T denoting the usual time-ordering operator. 7.8. Bang-bang
control
protocol
We propose to reduce the influence of the bfl-noise by applying to the qubit a continuous train of 7r-pulses along the crx-axis. This refocusing pulse scheme essentially corresponds to the standard quantum bang-bang procedure 179 ~ 181 or the Carr-Purcell-Gill-Meiboom echo technique from NMR 182 . For technical convenience, we consider the 7r-pulses to be of infinitesimal duration. This simplification is not crucial as described in 183 . The pulses are assumed to be separated by a constant time interval Tbb- The mean separation Tbfl between two bfl flips is assumed to be much longer than Tbb. For theoretical convenience, we also assume that Tbfl is shorter than the free precession period of the qubit. This too is not a crucial restriction. (It can always be overcome by changing to a co-precessing frame.)
82
0
0.02
0.04
0.06 t/T
0.08
0.1
. .Sys
Fig. 17. Schematic plot of a typical Poisonian bfl noise signal and its resulting random walk behavior (in the limit of small deviations). T h e periodic fast switching step function represents a bang-bang pulse with a time scale ratio: T^fl/rbb = 10 and yields a quite smaller random walk step-length. TS VS = , T denotes the evolution period of the qubit in the noiseless case.
Qualitatively, bang-bang control works as follows. Since rt,b
83
tCTz
Fig. 18. Schematic plot of noisy qubit evolution generated by Poissonian telegraph noise. The resulting random walk (dot-dashed line) on the Bloch sphere is comprised both of deviations Aod e p h in parallel to the free precession trajectory (dotted line), which correspond to dephasing, and deviations A
can be modelled as a discrete walk with steps that are randomly distributed in time, one step for each bfl flip (see e.g. 1 8 4 ). The average step length is essentially the product of the noise coupling strength a and the mean time the bfl. in its present state can influence the qubit. Without bang-bang control, this mean influence time is Tba, whereas with bang-bang control, it is reduced to i\>\>- Therefore, both with and without bang-bang control, the random walk has the same time distribution of steps, but with bang-bang control the step size can be significantly reduced roughly by a factor of the ratio of time scales Tbb/'Hjfl-
84
7.9. Random
Walk on the Bloch
sphere
Now we study this proposal quantitatively. We simulate these random walks both with and without bang-bang control by integrating both numerically and analytically the Schrodinger equation, Eq. (28), with the stochastic Hamiltonian of Eqs. (25-27). As generic conditions for the qubit dynamics, we choose eq = A q = do. Without loss of generality, we set the qubit's initial state to be spin-up along the 2-axis. If the qubit-bfl. coupling a were zero, then the qubit would simply precess freely on the Bloch sphere around the rotation axis a^ + a^ (the dotted line in Fig. 18). Hence, we expect for a sufficiently small coupling (a -C CIQ) only a slight deviation of the individual time evolution compared to the free evolution case (the dashed line in Fig. 18). For the coupling strength, we take a = 0.lCl0- All the following times and energies are given in units of the unperturbed system Hamiltonian, i.e., our time unit rs y s is given according the free precession time iTTsys/V^, and our energy unit is given by AE = we q + A q = y/2QgThe time scale ratio is taken to be Tbfl/rbb = 10 if not denoted otherwise. This approach accounts for the essential features of our specific situation: the long correlation time of the external noise, essentially Tbfi, its non-Gaussian statistics and its potentially large amplitude at low frequencies. These properties are crucial and are difficult, although not impossible, to take into account in standard master equation methods. 7.9.1. Numerical simulations We have numerically integrated Eq. (28) and averaged the deviations of the random walk evolution from the unperturbed trajectories for times up to 100rsyS over N = 103 realizations. Larger simulations have proven that convergence is already sufficient at this stage. We shall examine the rootmean-square (rms) deviations of this ensemble at given time points
Aarms(t) =
- £ (a](t) - Cis y ,i(0) \
(29)
3=1
with and without bang-bang control. In other approaches, such as those based on master equations, one separates dephasing and relaxation. Both are contained here in Eq. (29). We shall point out notable differences between these two channels. The deviation as a function of time is plotted in Fig. 19. The total deviations on intermediate time scales are suppressed by a
85 10"
X A + v*
numerical analytical bang-bang numerical bang-bang analytical
10"
< -10
Fig. 19. Time evolution of the rms deviations for bfl-induced random walks with and without bang-bang control at a coupling constant a = 0.1 and a typical flipping time scale Tbfl = 1 0 - 2 T s y s . The separation between two bang-bang pulses is Tbb = 1 0 _ 3 r g y s . The straight lines are square-root fits of the analytical derived random walk model variances (plotted as triangles). Inset: Components of the deviations from the free precession trajectory that are parallel t o it (dephasing) and perpendicular to it (relaxation/excitation) with bang-bang control.
ratio on the order of 10. A detailed numerical analysis shows that without bang-bang suppression, the deviations parallel to the free precession trajectory (which correspond to dephasing) are of similar size to those perpendicular to free precession (which correspond to relaxation/excitation). In contrast, with bang-bang control, dephasing is almost totally absent as one can see in the inset of Fig. 19. The main double-logarithmic plot of Fig. 19 shows that on short time scales (t ~ 0.1rsyS, which corresponds to ~ 10 random walk steps), deviations increase almost linearly in time. It is not until times on the order of rsys that the noise-induced deviations start to behave as typical classical random walks, increasing as a square-root in time.
86
7.9.2. Analytical random walk models We now develop analytical random walk models for our system. The random walk on the Bloch sphere is in general two-dimensional, consisting of both parallel and perpendicular deviations to the free evolution trajectory. Bangbang control, as was seen in the above numerical results and as will be seen in the following analytical results, essentially reduces the random walk to one-dimension as only the perpendicular deviations remain significant. In the following, we restrict ourselves to the long-time (many random walk steps) regime. We first calculate for both cases the probability distributions of the deviations after one bfl flip ("one-step deviations" in terms of the discrete random walk). The fluctuation of the period between r^, r leads to dephasing, which can be evaluated at a < e qi A q to A
dePh = 2TTCOS<j> ( —
q q
) Tbfl ~ ±2
\Tper
fper/
^q
2 a-r b f l .
i" e q
(30)
For the relaxation/excitation effect of the noise, one has to use the projection of the perturbation orthogonal to the free axis. In total, using T^,r ~ rper to first order in a, we find I
T
ACT^ = 27rcos^sin7?- 7 =cos0-^- ~ V2 V 2
Tper
e
A3 ^2\3/2arbfl-
( 31 )
(, q + ^ q j
The derivation of the maximal one-step deviation for the bang-bang controlled situation has to be handled differently. The deviation resulting from a bfl flip during a bang-bang pulse period is maximal if the step happens exactly at the moment of the second qubit spin-flip (i.e., in the middle of the bang-bang cycle). When this happens, the refocusing evolution has in its first half a drift, for example, to the "right" (compare Fig. 21) and in the last half an equal aberration. We average the maximal one-step deviation over one precession period in the usual rms manner to obtain sin^ X 4 a 2 r 4 d\ ^QTbb.
Obviously, this variance only contributes to relaxation.
(32)
87
Fig. 20. Plot of a typical one-step deviation from the unperturbed qubit trajectory with generic values for e q and A q . The fractions of the bfl fluctuations in avdirection have to be distinguished with respect to their effects on the qubit: those that yield dephasing deviations that are parallel to the free precession trajectory (proportional to sin
From the long-time limit of our analytical random walk distribution, we find for their variances in real space representation A<7bfi(7Vbfl) = 0 W
= x/iVbri^-arbfl
(33)
for the case without bang-bang control and Aahb(NhR) = ^ *
7
= y ^ W
(34)
for the case with it. In the large-A^bfl limit, this model shows excellent agreement with the numerical simulations.
Fig. 21. Sketch of a maximal one-step deviation during a bang-bang modulated cycle, which appears if the bfl state flips precisely at the intermediate bang-bang pulse time. The dephasing part of deviation evidently averages out, while a relaxating aberrance arise proportional to the noise-coupling constant a.
7.9.3. Bang-bang control working as a high-pass filter In order to measure the degree of noise suppression due to bang-bang control, we define the suppression factor Sta as follows for a given evolution time to <5t0(rbfl/rbb)
=
(35)
A^(to)'
We now systematically study the dependence of St0 on Tbfl/Tbb for a constant mean bfl switching rate Tbfl = 10 -2 r S y S at a fixed evolution time i 0 = r s y s . The numerical data in Fig. 22 show that the suppression efficiency is linear in the bang-bang repetition rate, 5TBys = /iTbn/Tbb- The numerically derived value of the coefficient, /inumerical ~ 1-679, is in excellent agreement with the analytical result ^analytical = \ / 5 / 2 ~ 1.581 from our saddle point approximation, Eqs. (33) and (34).
We have investigated the qubit errors that arise from the noise gener-
89
iol X Numerical result — linear fit •° 10
iooLr 10
_i
i
10J
10* 'bfl
10
10
/x bb
Fig. 22. The suppression factor <St0(Tbfl/Tt,b) = A ^{,^ a ^.°j evaluated for to = Ts y s as a function of the ratio of the mean switching time T\,n and the bang-bang pulse separation Tab-
ated by a single bistable fluctuator (bfl) in its semiclassical limit, where it behaves as a telegraph noise source. We numerically integrated a corresponding stochastic Schrodinger equation, Eq. (28), as well as analytically solved (in the long-time limit) appropriate random walk models. As a characteristic measure of the resulting dephasing and relaxation effects, we used the rms deviation of noisy evolutions compared to noiseless ones. To suppress the effects of this noise, we presented a bang-bang pulse sequence analogous to the familiar spin-echo method. We claimed this pulse sequence to be capable of refocusing most of the bfi-noise induced aberrations. Both in the case without bang-bang control and the case with it, there was excellent agreement between our numerical and analytical results on the relevant intermediate time scales (i.e., times after a short initial phase where deviations grow linearly instead of as a square-root in time, but before the qubit becomes totally decohered). Meanwhile, several other extensions of Ref. 185 - 183 have been proposed by other research groups. Ref. m includes a larger number of fluctuators,
90
described as semiclassical noise sources, but restricts itself to a single spinecho cycle. Ref. 170 analyzes extensively the importance of higher, nonGaussian cumulants and memory effects and arrives at a number of analytical results, but it does not treat the option of refocusing. Ref. 186 treats a full microscopic model and compares different variations of the bang-bang pulse sequence. Ref. 187 also treats a full microscopic model with potentially many fluctuators using a Lindblad-type approach and covers a wide range of ratios between the fluctuator and bang-bang pulse time scales. One of its main conclusions is that a Zeno effect is found in a parameter regime not covered by our work. Note that all of these other extensions of our work treat only the case of ideal bang-bang pulses. 8. Coupled qubits and beyond To implement a quantum algorithm, one must be able to entangle multiple qubits, so that an interaction term is required in the Hamiltonian describing a two qubit system. For two superconducting flux qubits, the natural interaction is between the magnetic fluxes. Placing the two qubits in proximity provides a permanent coupling through their mutual inductance 188 . Pulse sequences for generating entanglement have been derived for several superconducting qubits with fixed interaction energies 112>189. However, entangling operations can be much more efficient if the interaction can be varied and, ideally, turned off during parts of the manipulation. A variable coupling scheme for charge-based superconducting qubits with a bipolar interaction has been suggested recently 190 . For flux qubits, while switchable couplings have been proposed previously 73 ' 191 ) these approaches do not enable one to turn off the coupling entirely and require separate coupling and flux readout devices. As a new device, we propose a new coupling scheme for flux qubits in which the interaction is adjusted by changing a relatively small current. For suitable device parameters the sign of the coupling can also be changed, thus making it possible to null out the direct interaction between the flux qubits. Furthermore, the same device can be used both to vary the coupling and to read out the flux states of the qubits. We show explicitly how this variable qubit coupling can be combined with microwave pulses to perform the quantum Controlled-NOT (CNOT) logic gate. Using microwave pulses also for arbitrary single-qubit operations, this scheme provides all the necessary ingredients for implementation of scalable univeral quantum logic. The coupling is mediated by the circulating current J in a dc Supercon-
91
ducting QUantum Interference Device (SQUID), in the zero voltage state, which is coupled to each of two qubits through an identical mutual inductance Mqs [Fig. 23(a)]. A variation in the flux applied to the SQUID, $ s , changes J [Fig. 23(b)]. The response is governed by the screening parameter (3L = 2LI0/$0 and the bias current lb, where lb < / c ( $ s ) , the critical current for which the SQUID switches out of the zero voltage state at T = 0 in the absence of quantum tunneling. In flux qubit experiments 75 , the flux state is determined by a dc SQUID to which fast pulses of /;, are applied to measure IC($S,T). Thus, existing technology allows lb to be varied rapidly, and a single dc SQUID can be used both to measure the two qubits and to couple them together controllably.
(a) Inf ly . 1 .
x*x
l
I
(b)lr-T=5== • * / \ 0.5
o
:g=Lcxi:
3
So
* -M* &* J
-0.5
?
-1
-0.4 -0.2
0
0.2
0.4
Fig. 23. (a) SQUID-based coupling scheme. The admittance Y represents the SQUID bias circuitry, (b) Response of SQUID circulating current J to applied flux 3>s for /3L = 0.092 and / 6 / / c ( 0 . 4 5 * o ) = 0, 0.4, 0.6, 0.85 (top to bottom). Lower right inset shows J ( * s ) for same values of /;, near $ s = 0.45$o- Upper left inset shows Ic versus $ s .
The energy biases e° are determined by the flux bias of each qubit relative to 3>o/2. The tunnel frequencies Si/h are fixed by the device parameters, and are typically a few GHz. For two flux qubits, arranged so that a flux change in one qubit alters the flux in the other, the coupled-qubit Hamiltonian describing the dynamics in the complex 4-dimensional Hilbert space becomes H = 7ii® 7 ( 2 ) + 7 ( 1 ) ® H2 - (K/2)
(36)
where 1^ is the identity matrix for qubit i and K characterizes the coupling energy. For K < 0, the minimum energy configuration corresponds to antiparallel fluxes. For two flux qubits coupled through a mutual inductance Mqq, the interaction energy is fixed at KQ = -2M,,
r(l)
r(2)
92
For the configuration of Fig. 23(a), in addition to the direct coupling, KQ, the qubits interact by changing the current J in the SQUID. The response of J to a flux change depends strongly on If, [Fig. 23(b)]. When Iq ' switches direction, the flux coupled to the SQUID, A<J?i , induces a change A J in the circulating current in the SQUID, and alters the flux coupled from the SQUID to qubit 1. The corresponding coupling is Ks = J ^ A * ' 1 ) = - 2 M 2 S
IP
/(2)
Re(dJ/d$s)h-
(37)
The transfer function, (dJ/d<&s)i , is related to the dynamic impedance, Z, of the SQUID via 192 dJ/d$s = iuj/Z = l/L + iw/R,
(38)
where R is the dynamic resistance, determined by Y which dominates any loss in the Josephson junctions, and L is the dynamic inductance which, in general, differs from the geometrical inductance of the SQUID, L. We evaluate (dJ/d$s)I by current conservation, neglecting currents flowing through the junction resistances: Ib = Iy + 2I0 cos A7 sin 7 - 2C($O/2TT)
(39)
J = I0 cos 7 sin A7 - C ( $ 0 / 2 T T ) A 7 .
(40)
Here, Iy is the current flowing through the admittance Y{u>) [Fig. 23(a)], and IQ and C are the critical current and capacitance of each SQUID junction. The phase variables are related to the phases across each junction, 71 and 72, as A7 = (71 —72)/2 and 7 = (71+72V2. The phases are constrained by dA-y = (TT/$O )(<**» -
LdJ).
The expression for Ka in terms of Re (dJ/d$s)I (Eq. 37) requires the qubit frequencies to be much lower than the characteristic frequencies of the SQUID. This condition is satisfied by our choice of device parameters, and also ensures that the SQUID stays in its ground state during qubit entangling operations. Furthermore, it is a reasonable approximation to take the u> = 0 limit of Re (dJ/d$s)Ib to calculate Ks, so that we can solve Eqs. (39) and (40) numerically to obtain the working point; for the moment we assume Y(0) = 0. For the small deviations determining Ks, we linearize Eqs. (39) and (40) and solve for the real part of the transfer function in the low-frequency limit: / dJ \ _ 1 l-tan 2 A7tan 2 7 e V d $ J / b ~ 2 i ; i + 2l7(l-tan2A7tan27)'
( l
'
93
Here, we have introduced the Josephson inductance for one junction, Lj = $o/27r/ 0 cos A7COS7. For (jL > 1, Eq. (41) approaches 1/L, while for (3L
= (l/2Lj)(l
- t a n 2 A7tan27).
(42)
We see that Re (dJ/d^s)I becomes negative for sufficiently high values of lb and $ s , which increase 7 and A7. We choose the experimentally-accessible SQUID parameters L = 200 pH, C = 5 fF, and I0 = 0.48 //A, for which (3L = 0.092. The qubits are r(2) characterized by 7,(1) 0.46 /xA, M, s = 33 pH, and M„„ = 0.25 pH, yielding K0/h = -0.16 GHz. Choosing $ s = 0.45$ 0 , Eqs. (37) and (41) result in a net coupling strength K/h = (Ko + Ks)/h that is —0.3 GHz when Ib = 0, and zero when 7 b // c (0.45$o) = 0.57 [Fig. 24(a)]. The change in sign of Ks does not occur for all /?£. Figure 24(b) shows the highest achievable value of Ks versus 0L- We have adopted the optimal design at 13L = 0.092.
"0.6 Ib/Ic(0.45O0) Fig. 24. (a) Variation of K with /(, for $ s = 0.45$o and device parameters described in text, (b) Highest achievable value of Ks versus /3L evaluated at /;, = 0.85/ c (0.45$o); /o (and hence /3L) is varied for L = 200 pH.
We also need to consider crosstalk between the coupling and singlequbit terms in the Hamiltonian. When the coupling is switched, in addition to dJ/d$s being altered, J also changes, thus shifting the flux biases of the qubits. The calculated change in J as the coupler is switched from lb = 0 to 7b/7c(0.45$o) = 0.57 produces a change in the flux in each qubit corresponding to an energy shift 5e\/h = Sc^/h = 1.64 GHz. In addition, when the qubits are driven by microwaves to produce single-qubit rotations, the microwave flux may also couple to $ s . As a result, K is weakly modulated when the coupling would nominally be turned off. A typical microwave drive li(t)/h of amplitude 1 GHz results in a variation of about ±14 MHz about K = 0. When the bias current is increased to switch off the coupling, the SQUID
94
symmetry is broken and the qubits are coupled to the noise generated by the admittance Y. We estimate the decoherence due to this process using the technique outlined in section 7.3. We obtain as an intermeiate result J(W) = {fiMl/h) Im {dJ/d$s)h .
(43)
For the case F _ 1 = R, following the path to the static transfer function Eq. (41) and taking the imaginary part in the 1OW-/3L limit, we obtain lm(dJ/d$3)ib = —u/R = (w/4i?) tan 2 A 7 t a n 2 7 . Thus J(u>) = aco, where a = (MqSIq/4hR) tan 2 A7tan 2 7, and a(J& = 0) = 0. As h is increased to change the coupling strength, a increases monotonically. For the parameters described above and for R = 2.4 kfi, when the net coupling is zero [/(,//<;(0.45$o) = 0-57, Fig. 24(a)], we find a = 8 x 10~ 5 corresponding to a qubit dephasing time of about 500 ns, one order of magnitude larger than values currently measured in flux qubits 75 , however, shorter than the best values with flux echo. We now show that this configuration implements universal quantum logic efficiently. Any n-qubit quantum operation can be decomposed into combinations of two-qubit entangling gates, for example, CNOT, and singlequbit gates 193 . Single-qubit gates generate local unitary transformations in the complex 2-dimensional subspace for the corresponding individual qubit, while the two-qubit gates correspond to unitary transformations in the 4-dimensional Hilbert space. Two-qubit gates which cannot be decomposed into a product of single-qubit gates are said to be nonlocal, and may lead to entanglement between the two qubits 194 . Since we can adjust the qubit coupling K to zero, we can readily implement single-qubit gates with microwave pulses as described below. To implement the nonlocal two-qubit CNOT gate, we use the concept of local equivalence: the two-qubit gates U\ and U% are locally equivalent if U\ = k\U-2k2, wherefciand k^ are local two-qubit gates which are combinations of single-qubit gates applied simultaneously. These unitary transformations on the two single-qubit subspaces transform the gate U2 into U\. The local gate which precedes U2, £2, is given by £21 ® £22, where £21(22) is a single-qubit gate for qubit 1(2), while the local gate which follows U2, fci, is fcn ® £12, where fcn(i2) is a single-qubit gate for qubit 1(2) 195 . Our strategy is to find efficient implementation of a nonlocal quantum gate U2 that differs only by local gates, k\ and £2, from CNOT, using the methods in 194 , and the computational basis, in which the SQUID measures the projection of each qubit state vector onto the z-axis. The local equivalence classes of two-qubit operations have been shown
95
to be in one-to-one correspondence with points in a tetrahedron, the Weyl chamber. In this geometric representation, any two-qubit operation is associated with the point [c\, C2, C3], where CNOT corresponds to [n/2,0,0]. Furthermore, the nonlocal two-qubit gates generated by a Hamiltonian acting for time t can be mapped to a trajectory in this space 194 . If K is increased instantaneously to a constant value, the trajectory generated by Eq. (36) is well described by the following periodic curve [ci,c 2 ,c 3 ] = [Kvt/h,p\sinujt\
,p\sinu>t\].
(44)
Here, p is a function of the system parameters, v = e^e^/AEiAE^, and w = (AEi - AE2)/2H, where A £ ; = [(e°)2 + 5f]1'2 is the single-qubit energy level splitting. Independently of p, this trajectory reaches [7r/2,0,0] in a time tx = U-K/UJ when the coupling strength is tuned to K = Hu>/2nv, with n a nonzero integer. While this analytic solution contains the essential physics, it is an approximation and does not include vital experimental features, in particular, crosstalk and the finite rise time of the bias current pulse. To improve the accuracy, we perform a numerical optimization using Eq. (44) as a starting point, then add these corrections. We use tunnel frequencies Si/h = 5 GHz and 62/h = 3 GHz, and include the shifts of the single-qubit energy biases due to the crosstalk with Ks in Eq. (44) by adding a shift Sei proportional to K. We account for the rise and fall times of the current pulse by using pulse edges with 90% widths of 0.5 ns [see K{t) in Fig. 25]. We numerically optimize the variable parameters to minimize the Euclidean distance between the actual achieved gate and the desired Weyl chamber target CNOT gate. We find K/h = -0.30 GHz, el/h = 8.06 GHz, e°2/h = 2.03 GHz, and tK = 8.74 ns; tx is the time during which the qubit coupling is turned on. As outlined above, to achieve a true CNOT gate we still have to determine the pulse sequences which implement the requisite local gates that take this Weyl chamber target U2 to CNOT in the computational basis. Local gates may be implemented by applying microwave radiation, £*(£), which couples to ai , and is at or near resonance with the single-qubit energy level splitting AEi. We note that the single-qubit Hamiltonian driven by a resonant oscillating microwave field does not permit one to use standard NMR pulses, since the static and oscillating fields are not perpendicular, but rather are canted by an angle tan -1 ((5j/e°). To simplify the pulse sequence, we keep e\2 constant at the values used for the non-local gate generation. This imposes an additional constraint on the local gates: to generate a local two-qubit gate fci = fcn
96
ku and fci2 must be simultaneous and of equal duration. We satisfy this constraint by making the microwave pulse addressing one qubit resonant and that addressing the other slightly off-resonance. Using this offset and the relative amplitude and phase of the two microwave pulses as variables, we can achieve two different single-qubit gates simultaneously, leading to our required local two-qubit gate. The resulting pulse sequences for K and ii t 2 are shown in Fig. 25. The gate has a maximum deviation from CNOT in the computational basis of 1.6% in any matrix element. This error arises predominantly from the cross-coupling of the microwave signals for the two qubits and the weak modulation of the K = 0 state of the coupler during the single-qubit microwave manipulations. While small, this error could be reduced further by performing the numerical optimization with higher precision or by coupling the microwave flux selectively to each of the qubits and not to the SQUID. The total elapsed time of 29.35 ns is comparable to measured dephasing times in a single flux qubit 75 .
«r A
7.5 7 6.5 0
II Ml
--0.1 S-0.2 W-0.3 £ 1-51 S l cfO.5
;
10
LJ A
20
25
; 30
t(ns) Fig. 25. Pulse sequence for implementing CNOT gate. Energy scales in GHz. Total single-qubit energy bias €j(t) = e? + ei(t) + 6ei(t), where microwave pulses ei,2(t) produce single-qubit rotations in the decoupled configuration; crosstalk modulation of K{t) is shown (see text). The bias current is pulsed to turn on the interaction in the central region.
We have shown in the preceeding paragraphs that the inverse dynamic inductance of a dc SQUID with low f3i in the zero-voltage state can be varied by pulsing the bias current. This technique provides a variable-strength
97
X X X ' ^XyXJ ' X X X '
L
XjX J ' X X X 1 4 « | *
Fig. 26. Chain of flux qubits with intervening dc SQUIDs arranged to provide both variable nearest neighbor coupling and qubit readout.
interaction Ks between flux qubits coupled to the SQUID, and enables cancellation of the direct mutual inductive coupling K0 between the qubits so that the net coupling K can be switched from a substantial value to zero. By steering a nonlocal gate trajectory and combining it with local gates composed of simultaneous single-qubit rotations driven by resonant and off-resonant microwave pulses, we have shown that a simple pulse sequence containing a single switching of the flux coupling for fixed static flux biases results in a CNOT gate and full entanglement of two flux qubits on a timescale comparable to measured decoherence times for flux qubits. Furthermore, the same SQUID can be used to determine the flux state of the qubits. This approach should be readily scalable to larger numbers of qubits, as, for example, in Fig. 26. A wide range of other coupling schemes have been proposed and a few of them realized, which we will briefly review here. We will indicate a coupling Hamiltonian of the form Kaa ® a\, simply as AB. A constant ZZ coupling using a flux transformer has been demonstrated for flux qubits 188 . This is expcted to be possible up to strong coupling 196 . This transformer can in principle be made tunable by including a DC-SQUID loop 73 ' 74 or another superconducting switch 197 into the loop. Alternatively, the conjugate variable, charge, can be used in capacitive coupling, leading to a non-tunable XX + YY interaction 198 . All of these proposals need one physical coupling element per interacting pair, so they will conceivably be of nearest-neighbor type. For phase qubits, the coupling is not straightforwardly put into a Pauli matrix representation and the coupling matrix is generally bias dependent 199 and also depends on whether the qubits are encoded into adiabatic eigenstates or a global basis; however, the resulting interaction certainly has entangling power 189 . A spectroscopic hint on two-qubit entanglement 200 and three qubit interaction 201 has been observed. Very recently, coherent manipulations have been achieved in this system in a setup which allows for simultaneous measurement 113 . In the charge qubit case, a resonator coil between all qubits has been pro-
98
posed for a YY interaction 30 , which is tunable and of arbitrary range but does not appear to be compatible with parallel operation. Note, that a large coil usually provides strong coupling to external noise, but it can in principle be replaced by a large classical Josephson junction using the kinetic inductance 2 0 2 _ 2 0 4 . Using the screening currents inside loop-shaped charge qubits, a tunable XX nearest-neighbor interaction can be implemented 205 . More easily, one can implement a capacitive interaction, which leads to ZZ. This is typically constant; however, using the capacitance of additional Josephson junctions it can be made tunable 190 . Coherent charge dynamics and conditional operations with constant capacitive interaction have been demonstrated 79>112. A tunable nearest-neighbor XX + YY coupling is achieved by connection with SQUIDs; however, the interaction strength is rather low 206 . Charge qubits can profit from a cavity, mediating an XXinteraction 207 and nonclassical radiation 208 . The interaction between quantronia can be implemented using a capacitor and is expected to be of XX + YY structure. This variety of possible interactions opens the question, which type of interactions are efficient and about the importance of tunability, long range, and parallel operation. Clearly, tunability is not needed for the demonstration of gates on a few qubits similar to NMR 209>10»112.210; however, with an increasing number of qubits, controls are supposed to become more and more complex. In fact, the in situ tunability of interactions is one of the main advantages of superconducting qubits and shouldnot be given up. On the other hand, long range interactions are less crucial as algorithms operate efficiently even on linear nearest neighbor arrays 211 . It is on the other hand absolutely crucial to operate in parallel. 9. Summary Let's take an overview of the present status of superconducting Josephson qubits from a view point of the theme of this conference ; "Are the DiVincenzo criteria satisfied by today's experiments ?" (1) A scalable physical system of well-characterized qubits Employing the very advanced micro and nano fabrication technology from the semiconductor industry, superconducting qubits have strong potential advantage in scalability towards the circuit level, including tunable coupling switches.
99
(2) The ability to initialize the state of the qubits to a simple fiducial state The requirement of proper initialization can be satisfied if superconducting qubit devices are operated at low enough temperature ~20 mK, which is also far below the critical temperature of aluminum (~1.2 K), niobium (~9 K) or other superconducting materials, such that the ground state is occupied with probability nearly 1. The qubit state is protected from rapid energy relaxation once it is cooled below the critical temperature, because it is disconnected from the phonon bath and there is an energy gap in the quasi-particle density of states. (3) Long (relative) decoherence times, much longer than the gate-operation time A typical decoherence time of a present superconducting Josephson qubit is T?| «0.5/xs during free induction decay 81 ' 97 ' 212 and it reaches T2 «4/zs under the echo pulse technique 91 . Whereas a typical time for single qubit gate operation, e.g a 7r-rotation pulse, is an order of 0.1 ns under strong driving. The present few ns 7r-rotation time via the conditional side-band transition of qubit LC-resonator coupled system 114,118 is ready to improve as a design enabling strong driving. In order to demonstrate C-NOT gate operation, a series of phase-shifted composite side-band pulses are needed. So, the present typical decoherence time normalized by a single two-qubit gate operation time would be an order of 103. This experimental value is a result at the optimal operating point, which is a commonly used strategy to decouple qubit to the first order from outside "charge" and "flux" noise 81 . (4) A universal set of quantum gates An arbitrary rotational gate for a single qubit operation is realized by applying resonant microwave pulses from an on-chip microwave circuit. This is already demonstrated and established in all types of superconducting qubits, i.e., the "charge" 77 , "phase" 69 ' 70 , "charge-phase" 81,109 , and "flux"75 qubits. In order to achieve noise torelant qubit operation and to achieve two-axis qubit rotation without using slow detuning technique, NMR-like multi pulse sequence control has been demonstrated in a "charge-phase" qubit 95 and also in a "flux"96 qubit re-
100
spectively. The antiphase oscillation of the |01) and |10) states in the capacitively coupled Josephson phase two-qubits 113 has been observed. So far, the C-NOT gate operation has been demonstrated only in a "charge" two-qubit system 112 . (5) A qubit-specific measurement capability Measurement of qubit state can be done, for example, using an extremely sensitive magnetic flux quantum detector namely a SQUID which is placed adjacent to the qubit. Single-shot readout has been achieved in a "flux" qubit for the first time in the slow ramping readout method 213 ' 105 and recently in a "charge" qubit using an S-SET pulsed readout technique 107 and very recently in a "charge-phase" qubit coupled to a transmittion line resonator using reflection wave phase detection 109 . Furthermore, the state of the qubit can be measured by dispersive technique that probes the second derivative of the state energy with respect to qubit bias parameters. The phase degree of freedom is used to perform inductive readout 111 using Josephson bifurcation amplification 110,212 . This novel readout projects the state of the qubit typically in a few hundred nano second which is roughly two orders of magnitude faster than the conventional switching readout which inevitably includes the dead time due to a large numbers of generated quasiparticles to relax to the ground state. This new readout method has a lot of advantages with an improved signal to noise ratio and contrast. Thus the readout visibility of the coherent oscillation is becoming better approaching the unit visibility.
Certainly, theoretical proposals for meeting all DiVincenzo's criteria have been put forward and no principal obstacle is known. However, a few experimental issues in decoherence, in particular the loss of visibility, are not fully understood today and may prove problematic if they are fundamental. Acknowledgments We would like to thank Prof. M. Nakahara for organizing this fascinating conference and Kinki University for supporting it. We would like to thank all co-workers with whom we have obtained the results presented here. FKWs work is supported by ARDA through ARO contract No. P-43385-PH-QC
101 and D F G through SFB 631.
References 1. G. Moore, Electronics 38, 114 (1965). 2. G. Moore, No Exponential is forever, Keynote Presentation International Solid State Circuits Conference, 2003. 3. I. Co., Expanding Moore's law, Available at ftp://download.intel.com/labs/eml/download/EML_opportunity.pdf (2002). 4. M. Nielsen and I. Chuang, Quantum Computation and Quantum Information (Cambridge University Press, Cambridge, UK, 2000). 5. J. Preskill, Quantum computation lecture notes, Available at http://www.theory.caltech.edu/people/preskill/ph219/. 6. D. DiVincenzo, Science 270, 255 (1995). 7. D. DiVincenzo, Fortschr. Phys. 48, 771 (2000). 8. R. Hughes, the ARDA Quantum Information Science, and T. Panel, A Quantum Information Science and Technology Roadmap, available at http://qist.lanl.gov. 9. W. Kaminsky, S. Lloyd, and T. Orlando, quant-ph/0403090 (unpublished). 10. L. Vandersypen et al, Nature 414, 883 (2001). 11. N. Gershenfeld and I. Chuang, Science 275, 350 (1997). 12. C. Monroe et al., Phys. Rev. Lett. 75, 4714 (1995). 13. J. Cirac and P. Zoller, Phys. Rev. Lett. 74, 4091 (1995). 14. Q. Turchette et al, Phys. Rev. Lett. 75, 4710 (1995). 15. T. Pellizzari, S. Gardiner, J. Cirac, and P. Zoller, Phys. Rev. Lett. 75, 3788 (1995). 16. E. Knill, R. Laflamme, and G. Milburn, Nature 409, 46 (2001). 17. H.-J. Briegel et al., Journal of modern optics 47, 415 (2000). 18. D. Kielpinski, C. Monroe, and D. J. Wineland, Nature 417, 709 (2002). 19. S. Sze, Semiconductor devices, physics and technology (Wiley-VCH, New York, 2001). 20. S. Thompson et al, Intel Technology Journal 6, (2002). 21. Y. Imry, Introduction to Mesoscopic Physics (Oxford University Press, Oxford, 1997). 22. Mesoscopic electron transport, No. E345 in Proceedings of the NATO Advanced Study Institute, edited by L. Kouwenhoven, G. Schon, and L. Sohn (Kluwer, Dordrecht, 1997). 23. A. Leggett, Progr. Theor. Phys. Suppl. 69, 80 (1980). 24. A. Leggett and A. Garg, Phys. Rev. Lett. 54, 857 (1985). 25. C. Tesche, Phys. Rev. Lett. 64, 2358 (1990). 26. K. Likharev, Usp. Fiz. Nauk. 139, 169 (1983). 27. P. Anderson, in Lectures on the Many-Body Problem, edited by E. Caianiello (Academic Press, New York, 1964), p. 113. 28. Single Charge Tunneling, No. B294 in NATO ASI Series, edited by H. Grabert and M. Devoret (Plenum, New York, 1992).
102 29. V. Bouchiat, D. V. et P. Joyez, D. Esteve, and M. Devoret, Physica Scripta T76, 165 (1998). 30. Y. Makhlin, G. Schon, and A. Shnirman, Rev. Mod. Phys. 73, 357 (2001). 31. B. Kane, Nature 393, 133 (1998). 32. D. Loss and D. DiVincenzo, Phys. Rev. A 57, 120 (1998). 33. J. Elzerman et al., Phys. Rev. B 67, 161308 (2003). 34. R. Clark et al, Los Alamos Science 27, 284 (2002). 35. L. Vandersypen et al., in Quantum Computing and Quantum Bits in Mesoscopic Systems (Kluwer, Dordrecht, 2002), Chap. 22. 36. G. Burkhard, H. Engel, and D. Loss, Fortschr. Physik 48, 965 (2000). 37. R. Hanson et al., Phys. Rev. Lett. 94, 196802 (2005). 38. H. K. Onnes, Leiden Comm. 122b, 124 (1911). 39. J. Bardeen, L. Cooper, and J. Schrieffer, Phys. Rev. 108, 1175 (1957). 40. M. Tinkham, Introduction to Superconductivity (McGraw-Hill, New York, 1996). 41. P. de Gennes, Superconductivity of metals and alloys (Benjamin, N.Y., 1966). 42. J. Schrieffer, Theory of Superconductivity (Cummings, N.Y., 1988). 43. F. London, Superfluids (Wiley and Sons, N.Y., 1954). 44. R. Doll and M. Nabauer, Phys. Rev. Lett. 7, 51 (1961). 45. L. Tian et al, in Quantum Mesoscopic Phenomena and Mesoscopic Devices in Microelectronics (Kluwer, Dordrecht, 2000), Chap. Decoherence of the superconducting persistent current qubit, edited by I. O. Kulik and R. Ellialogulu. 46. L. Ioffe, V. Geshkenbein, C. Helm, and G. Blatter, Phys. Rev. Lett. 93, 057001 (2004). 47. B. Josephson, Phys. Lett. 1, 251 (1962). 48. A. Barone and G. Paterno, Physics and applications of the Josephson effect (Wiley, N.Y., 1982). 49. K. Likharev, Dynamics of Josephson junctions and circuits (Gordon and Breach, New York, 1986). 50. A. Golubov, M. Kupriyanov, and E. IFichev, Rev. Mod. Phys. 76, 411 (2004). 51. R. Feynman, Lectures in Physics (Addison Wesley, New York, 1970), Vol. 3. 52. L. Landau and E. Lifshitz, Statistical Physics 2 (Akademie-Verlag, Berlin, 1979). 53. S. Pereverzev et al, Nature 388, 449 (1997). 54. K. Sukhatme, Y. Mukharsky, T. Chui, and D. Pearson, Nature 411, 280 (2001). 55. B. P. Anderson and M. A. Kasevich, Science 282, 1686 (1998). 56. P. Hadley, private communication. 57. T. Orlando and K. Delin, Foundations of applied superconductivity (Prentice Hall, New York, 1991). 58. T. van Duzer and C. Turner, Principles of superconductive devices and circuits (Elsevier, Amsterdam, 1999). 59. V. Ambegaokar, U. Eckern, and G. Schon, Phys. Rev. Lett. 48, 1745 (1982).
103 60. J. Clarke, in Applications of Superconductivity (Kluwer, Dordrecht, 2000), Chap. Low- and High-Tc SQUIDs and Some Applications, p. 1, edited by H. Weinstock. 61. J. Niemeyer, Supercond. Sci. and Technology 13, 546 (2000). 62. J. von Delft and H. Schoeller, Ann. Phys. (Leipzig) 7, 225 (1998). 63. A. Zaikin and G. Schon, Phys. Rep. 198, 237 (1990). 64. H. van der Zant and R. Fazio, Phys. Rep. 355, 236 (2001). 65. R. Feynman and F. Vernon, Ann. Phys. (N.Y.) 24, 118 (1963). 66. G. Ingold, in Quantum transport and dissipation (Wiley-VCH, Weinheim, 1998), Chap. Dissipative quantum systems. 67. A. Caldeira and A. Leggett, Phys. Rev. Lett. 46, 211 (1981). 68. A. Caldeira and A. Leggett, Ann. Phys. (NY) 149, 374 (1983). 69. J. Martinis, S. Nam, J. Aumentado, and C. Urbina, Phys. Rev. Lett 89, 117907 (2002). 70. Y. Yu et ai, Science 286, 889 (2002). 71. A. Berkley et ai, Science 300, 1548 (2003). 72. J. Friedman et ai, Nature 46, 43 (2000). 73. J. Mooij et ai, Science 285, 1036 (1999). 74. T. P. Orlando et ai, Phys. Rev. B 60, 15398 (1999). 75. I. Chiorescu, Y. Nakamura, C. Harmans, and J. Mooij, Science 299, 1869 (2003). 76. E. Il'ichev et ai, Phys. Rev. Lett. 91, 097906 (2003). 77. Y. Nakamura, Y. Pashkin, and J. Tsai, Nature 398, 786 (1999). 78. Y. Nakamura, Y. A. Pashkin, T. Yamamoto, and J. S. Tsai, Phys. Rev. Lett. 88, 047901 (2002). 79. Y. Pashkin et ai, Nature 421, 823 (2003). 80. K. Lehnert et ai, Phys. Rev. Lett. 90, 027002 (2003). 81. D. Vion et a/., Science 296, 286 (2002). 82. L. B. Ioffe et ai, Nature 398, 679 (1999). 83. A. Blais and A. Zagoskin, Phys. Rev. A 61, 042308 (2000). 84. L. B. Ioffe et ai, Nature 398, 678 (2002). 85. G. J. Dolan, Appl. Phys. Lett. 31, 3 (1977). 86. A. Leggett, J. Phys. C 14, 415 (2002). 87. W. Diir, C. Simon, and J. Cirac, Phys. Rev. Lett. 89, 210402 (2002). 88. D. Crankshaw and T. Orlando, IEEE Trans. Appls. Supercond. 11, 1006 (2001). 89. A. van den Brink, Phys. Rev. B 7 1 , 064503 (2005). 90. G. Burkard et al, Phys. Rev. B 7 1 , 134504 (2005). 91. P. Bertet et al, cond-mat/0412485 (unpublished). 92. A. Leggett et al., Rev. Mod. Phys. 59, 1 (1987). 93. Y. Nakamura, Y. Pashkin, and J. Tsai, Phys. Rev. Lett. 87, 246601 (2001). 94. J. Clarke et al., Science 239, 992 (1988). 95. E. Collin et ai, Phys. Rev. Lett. 93, 157005 (2004). 96. T. Kutsuzawa et al., Appl. Phys. Lett. 87, 037501 (2005). 97. K. Harrabi, F. Yoshiwara, Y. Nakamura, and J. S. Tsai, 1T1696, poster presentation PB-F-92 in the 24th International Conference on Low Tern-
104 perature Physics (LT24), Orlando, Florida, USA, (2005). (unpublished). 98. K. Kakuyanagi, S. Saito, K. Semba, and H. Takayanagi, to be presented in the Quantum Information Technology Symposium (QIT13), Sendai, Japan, (2005). (unpublished). 99. R. Simmonds et al., Phys. Rev. Lett. 93, 077003 (2004). 100. K. B. Cooper et al., Phys. Rev. Lett. 93, 180401 (2004). 101. T. Bauch et al., Phys. Rev. Lett. 94, 087003 (2005). 102. K. Inomata et al., Phys. Rev. Lett. 95, 107005 (2005). 103. K. Ueda et al, Appl. Phy. Lett. 86, 172502 (2005). 104. H. Shimakage, K. Tsujimoto, Z. Wang, and M. Tonouchi, Appl. Phys. Lett. 86, 072512 (2005). 105. H. Tanaka et al., cond-mat/0407299 (unpublished). 106. H. Nakano et al., cond-mat/0406622 (unpublished). 107. O. Astafiev et al, Phys. Rev. B 69, 180507(R) (2004). 108. P. Bertet et al, Phys. Rev. B 70, 100501(R) (2004). 109. A. Wallraff et al, Phys. Rev. Lett. 95, 060501 (2005). 110. I. Siddiqi et al., Phys. Rev. Lett. 94, 027005 (2005). 111. A. Lupascu et al., Phys. Rev. Lett. 93, 177006 (2004). 112. T. Yamamoto et al., Nature 425, 941 (2003). 113. R. McDermott et al., Science 307, 1299 (2005). 114. I. Chiorescu et al., Nature 431, 159 (2004). 115. A. Wallraff et al., Nature 431, 162 (2004). 116. D. I. Schuster et al, Phys. Rev. Lett. 94, 123602 (2005). 117. J. Johansson et al., cond-mat/0510457 (unpublished). 118. K. Semba et al, 1T1271, poster presentation PB-F-86 in the 24th International Conference on Low Temperature Physics (LT24), Orlando, Florida, USA, (2005). (unpublished). 119. E. Hecht, Optics (Addison-Wesley, Reading, Mass., 1987). 120. A. Peres, Quantum Theory: Concept and Methods (Kluwer, Dordrecht, 1993). 121. D. Giulini et al., Decoherence and the Appearance of a Classical World in Quantum Theory (Springer, Heidelberg, 1996). 122. W. Unruh, Phys. Rev. D 40, 1053 (1989). 123. C. Cohen-Tannoudji, J. Dupont-Roc, and G. Grynberg, Atom-Photon Interactions (Wiley Interscience, New York, 1998). 124. E. Schrodinger, Naturwissenschaften 23, 807, 823, 844 (1935). 125. V. Emery and A. Luther, Phys. Rev. B 9, 215 (1974). 126. P. Anderson, Phys. Rev. Lett. 18, 1049 (67). 127. R. J. Schoelkopf et al, in Quantum Noise, Nato ASI (Kluwer, Dordrecht, 2002), Chap. Qubits as spectrometers of quantum noise, edited by Yu.V. Nazarov and Ya.M. Blanter. 128. R. Aguado and L. Kouwenhoven, Phys. Rev. Lett. 84, 1986 (2000). 129. A. Cottet, Ph.D. thesis, Universite Paris 6, 2002. 130. J. Martinis et al., Phys. Rev. B 67, 094510 (2003). 131. C. Slichter, Principles of magnetic resonance, No. 1 in Springer Series in Solid-State Sciences (Springer, Berlin, 1996).
105
132. 133. 134. 135. 136. 137. 138. 139. 140. 141. 142. 143. 144. 145.
146. 147. 148. 149. 150. 151. 152. 153. 154. 155. 156. 157. 158. 159. 160. 161. 162. 163. 164. 165. 166.
M. Grifoni, E. Paladino, and U. Weiss, Eur. Phys. J B 10, 719 (1999). L. Tian, S. Lloyd, and T. Orlando, Phys. Rev. B 65, 144516 (2002). A. Garg, J. Onuchic, and V. Ambegaokar, J. Chem. Phys. 83, 4491 (1985). A. Abragam, Principles of nuclear magnetism, Vol. 32 of International series of monographs on physics (Clarendon Press, Oxford, 1983). S. Adler, Studies in history and philosophy of modern physics 34, 135 (2003). S. Nakaijma, Progr. Theor. Phys. 20, 948 (1958). R. Zwanzig, Phys. Rev. 124, 983 (1961). P. Argyres and P. Kelley, Phys. Rev. 134, A98 (1964). U. Weiss, Quantum Dissipative Systems, No. 10 in Series in modern condensed matter physics, 2 ed. (World Scientific, Singapore, 1999). H. Spohn, Rev. Mod. Phys. 52, 569 (1980). G. Lindblad, Commun. Math. Phys. 48, 119 (1976). R. Alicki and K. Lendi, Quantum dynamical semigroups and applications, No. 286 in Lecture notes in physics (Springer, Berlin, 1976). H. Charmichael, An open systems approach to quantum optics (Springer, Berlin, 1993). C. Gardiner and P. Zoller, Quantum Noise. A Handbook of Markovian and Non-Markovian Quantum Stochastic Methods with Applications to Quantum Optics, Springer series in synergetics (Springer, Berlin, 1999). D. Loss and D. DiVincenzo, cond-mat/0304118 (unpublished). F. Bloch, Phys. Rev. 105, 1206 (1957). A. Redfield, IBM J Res. Develop. 1, 19 (1957). M. Plenio and P. Knight, Rev. Mod. Phys. 70, 101 (1998). J. Dalibard, Y. Castin, and K. Molmer, Phys. Rev. Lett. 68, 580 (1992). A. Korotkov, Phys. Rev. B 67, 235408 (2003). T. Costi, Phys. Rev. Lett. 80, 1038 (1998). S. Kehrein and A. Mielke, J. Stat. Phys. 90, 889 (1998). M. Keil and H. Schoeller, Phys. Rev. B 63, 180302 (2001). A. Leggett, Phys. Rev. B 30, 1208 (1984). C. van der Wal, F. Wilhelm, C. Harmans, and J. Mooij, Eur. Phys. J. B 31, 111 (2003). T. Robertson et al, Phys. Rev. B 72, 024513 (2005). F. K. Wilhelm, Phys. Rev. B 68, 060503(R) (2003). S. K. and. S. Kehrein and J. von Delft, Phys. Rev. B 70, 014516 (2004). F. Wilhelm, S. Kleff, and J. van Delft, Chem. Phys. 296, 345 (2004). M. Thorwart, E. Paladino, and M. Grifoni, Chem. Phys. 296, 333 (2004). M. Goorden, M. Thorwart, and M. Grifoni, Phys. Rev. Lett. 93, 267005 (2004). A. Shnirman and G. Schoen, Phys. Rev. Lett. 57, 15400 (1998). C. van der Wal et al., Science 290, 773 (2000). P. Joyez et al, Journal of superconductivity 12, 757 (1999). W. Coffey, Y. Kalmykov, and J. Waldron, The Langevin equation: With applications in physics chemistry and electrrical engineering, No. 11 in Series in contemporary chemical physics (World Scientific, Singapore, 1996).
106
167. N. Prokovev and P. Stamp, Rep. Prog. Phys. 63, 669 (2000). 168. H. Gassmann, F. Marquardt, and C. Bruder, Phys. Rev. E 66, 041111 (02). 169. E. Paladino, L. Faoro, G. Falci, and R. Fazio, Phys. Rev. Lett. 88, 228304 (2002). 170. K. Rabenstein, V. Sverdlov, and D. Averin, J E T P Lett. 79, 646 (2004). 171. Y. Galperin, B. Altshuler, and D. Shantsev, in Fundamental problems of mesoscopic physics: Interaction and decoherence, NATO-ASI (Plenum, New York, 2003), Chap. Low-frequency noise as a source of dephasing a qubit. 172. A. Shnirman, Y. Makhlin, and G. Schon, Phys. Scr. T102, 147 (2002). 173. P. Dutta and P. Horn, Rev. Mod. Phys. 53, 497 (1981). 174. R. Wakai and D. van Harlingen, Phys. Rev. Lett. 58, 1687 (1987). 175. N. Zimmermann, J. Cobb, and A. Clark, Phys. Rev. B 56, 7675 (1997). 176. M. Weissman, Rev. Mod. Phys. 60, 537 (1988). 177. N. van Kampen, Stochastic processes in physics and chemistry (Elsevier, Amsterdam, 1997). 178. L. Arnold, Stochastische Differentialgleichungen (Oldenbourg, Munich, 1973). 179. L. Viola and S. Lloyd, Phys. Rev. A 58, 2733 (1998). 180. L. Viola, E. Knill, and S. Lloyd, Phys. Rev. Lett. 82, 2417 (1999). 181. L. Viola, S. Lloyd, and E. Knill, Phys. Rev. Lett. 83, 4888 (1999). 182. H. Carr and E. Purcell, Phys. Rev. 94, 630 (1954). 183. H. Gutmann, F. Wilhelm, W. Kaminsky, and S. Lloyd, Quant. Inf. Proc. 3, 247 (2004). 184. G. Weiss, Aspects and Applications of the Random Walk (North-Holland, Amsterdam, 1994). 185. H. Gutmann, W. Kaminsky, S. Lloyd, and F. Wilhelm, Phys. Rev. A 7 1 , 020302(R) (2005). 186. L. Faoro and L. Viola, Phys. Rev. Lett. 92, 117905 (2004). 187. G. Falci, A. D'Arrigo, A. Mastellone, and E. Paladino, Phys. Rev. A 70, R40101 (2004). 188. J. Majer et al., Phys. Rev. Lett. 94, 090501 (2005). 189. F. Strauch et al, Phys. Rev. Lett. 9 1 , 167005 (2003). 190. D. Averin and C. Bruder, Phys. Rev. Lett. 9 1 , 057003 (2003). 191. J. Clarke et al., Phys. Scr. T102, 173 (2002). 192. C. Hilbert and J. Clarke, J. Low Temp. Phys. 6 1 , 237 (1985). 193. A. Barenco, D. Deutsch, A. Ekert, and R. Josza, Phys. Rev. Lett. 74, 4083 (1995). 194. J. Zhang, J. Vala, S. Sastry, and K. Whaley, Phys. Rev. A 67, 042313 (2003). 195. Y. Makhlin, Quant. Inf. Proc. 1, 243 (2002). 196. J. You, Y. Nakamura, and F. Nori, Phys. Rev. B 7 1 , 0242532 (2005). 197. M. Storcz and F. Wilhelm, Appl. Phys. Lett. 83, 2389 (2003). 198. L. Levitov, T. Orlando, J. Majer, and J. Mooij, cond-mat/0108266 (unpublished) . 199. P. Johnson et al., Phys. Rev. B 67, 020509 (2002). 200. A. Berkley et al., Science 300, 1548 (2003). 201. H. Xu et al., Phys. Rev. Lett 94, 027003 (2005).
107 202. 203. 204. 205. 206. 207. 208. 209. 210. 211. 212. 213.
J. You, J. Tsai, and F. Nori, Phys. Rev. Lett. 89, 197902 (2002). L. Wei, Y. x. Liu, and F. Nori, Europhys. Lett. 67, 1004 (2004). L. Wei, Y. x. Liu, and F. Nori, Phys. Rev. B 7 1 , 134506 (2005). J. Lantz, M. Wallquist, V. Shumeiko, and G. Wendin, Phys. Rev. B 70, 140507(R) (2004). J. Siewert, R. Fazio, G. Palma, and E. Sciacca, J. Low Temp. Phys. 118, 795 (2000). Y. x. Liu, L. Wei, and F. Nori, Phys. Rev. A 7 1 , 063820 (2005). M. Paternostro, G. Falci, M. Kim, and G. Palma, Phys. Rev. B 69, 214502 (2004). A. Spoerl et al., quant-ph/0504202 (unpublished). C. Rigetti, A. Blais, and M. Devoret, Phys. Rev. Lett. 94, 240502 (2005). A. Fowler, S. Devitt, and L. Hollenberg, Quant. Inf. Comp. 4, 237 (2004). I. Siddiqi et al, cond-mat/0507548 (unpublished). H. Tanaka, Y. Sekine, S. Saito, and H. Takayanagi, Physica C 368, 300 (2002).
108
Controlling Three Atomic Qubits H. Haffner 1 ' 2 , M. Riebe 1 , F. Schmidt-Kaler 1 , W. Hansel 1 , C. Roos 1 , M. Chwalla 1 , J. Benhelm 1 , T. Korber 1 , G. Lancaster 1 , C. Becher 1 , D.F.V. James 3 and R. Blatt 1 ' 2 1
2
3
Institut fur Experimentalphysik,
Institut fur Quantenoptik
Theoretical Division
Innsbruck,
Austria
und Quanteninformation, Osterreichische Wissenschaften, Austria
Akademie
der
T~4, Los Alamos National Laboratory, Los Alamos NM 87545, United States
We present a series of experiments where up to three ions held in a Paul trap are entangled, a given number of ions is selectively read out while conditional single-quantum-bit (qubit) operations are performed coherently on the remaining ion(s). Using these techniques, we demonstrate also a state transfer of a quantum bit from one ion to another one using two measurements and entanglement between an auxiliary ion and the target ion — also known as teleportation.
1. Introduction Quantum information processing rests on the ability to control a quantum register 1 . In particular this includes initialization, manipulation and read-out of a set of qubits. After initialization, a sequence of quantum gate operations implements the algorithm, which usually generates multipartite entangled states of the quantum register. Finally, the outcome of the computation is obtained by measuring the state of the individual quantum bits. In addition, for some important algorithms such as quantum error correction 1 _ 5 and teleportation 6 a subset of the quantum register is read out selectively and subsequently operations on other qubits are carried out conditioned on the measurement result*.
'Indeed, such an error-correction scheme has been carried out recently in an ion trap by Chiaverini and co-workers 7 in an ion trap.
109
This idea of the selective read-out of a quantum register gains some additional appeal when carried out on an entangled register, because there the measurement process can be demonstrated in an extraordinary clear way. Producing entangled states is also the key ingredient for quantum information processing, and last but not least, such experiments realize some of the Gedanken experiments which helped significantly to develop quantum mechanics. Creation of entanglement with two or more qubits has already been demonstrated in the references 8~"14. However, so far only trapped ions have allowed to create entanglement in a completely deterministic way 9 . Our experiment allows the deterministic generation of 3-qubit entangled states and the selective read-out of an individual qubit followed by local quantum operations conditioned on the read-out. As we will see later, the selective read-out of the quantum register illuminates the measurement process in a very clear way. The paper is organized as follows: first the deterministic creation of maximally entangled three-qubit states, specifically the Greenberger-HorneZeilinger (GHZ) state and the W-state, with a trapped-ion quantum computer is discussed15. In sections 4 and 5 we show how the qubits can be read out selectively and how GHZ- and W-states are affected by such local measurements. Next we demonstrate in section 6 operations conditioned on the read-out. This enables us to transform tripartite entanglement deterministically into bipartite entanglement with local operations and measurements. It also realizes a quantum eraser along the lines proposed in Ref.16. Finally, we implement a full deterministic quantum teleportation on demand 17 (see section 7). 2. Experimental setup All experiments are performed in an elementary ion-trap quantum processor 18 ' 19 . In order to investigate tripartite entanglement 20 ^ 22 , we trap three 4 0 Ca + ions in a linear Paul trap where they arrange themselves in a linear chain with an inter-ion distance of 5 ^m. Qubits are encoded in a superposition of the S1/2 ground state and the metastable D 5 / 2 state (lifetime r ~ 1.16 s). Each ion-qubit is individually manipulated by a series of laser pulses on the S = S1/2 (mj=-l/2) to D = D 5 / 2 (nij=-l/2) quadrupole transition near 729 nm employing narrowband laser radiation tightly focused onto individual ions in the string. The entire quantum register is prepared by Doppler cooling, followed by sideband ground state cooling of the center-of-mass vibrational mode (w = 2ir 1.2 MHz) as required for the controlled interaction of the ions according to the original proposal by Cirac
110
and Zoller23. Finally, the ions' electronic qubit states are initialised in the S-state by optical pumping. The operations which modify individual qubits and simultaneously connect a qubit to the bus (the center-of-mass mode) are performed by applying laser pulses on the carrier t or the "blue" sideband * of the S—>D transition. Qubit rotations can be written as unitary operations in the following way (c.f. 2 4 ): carrier rotations are given by Rc(6,
(1)
whereas transitions on the blue sideband are denoted as R+(9,(f)
=exp i - (e^a+af +
e-^a'a)
(2)
Here a^ are the atomic raising and lowering operators which act on the electronic quantum state of an ion by inducing transitions from the \S) to \D) state and vice versa (notation: er+ = \D)(S\). The operators a and a* denote the annihilation and creation of a phonon at the trap frequency u>, i.e. they act on the motional quantum state. The parameter 6 depends on the strength and the duration of the applied pulse and ip the relative phase between the optical field and the atomic polarization. Ions are numbered in analogy to binary numbers such that the first ion is the right-most with the least significance. Defining the D-Level as logical 0, we obtain the following ordering of the basis: \DDD), \DDS), \DSD).... 3. Preparing G H Z - and W - s t a t e s Entanglement of three qubits can be divided into two distinct classes21: GHZ-states and W-states. Choosing one representative of each class (\GHZ) = (\SSS) + \DDD))/y/2 and \W) = {\DDS) + \DSD) + \SDD))/^) any pure entangled three qubit state can be created by single qubit operations on either \GHZ) or \W). We synthesize GHZ-states using a sequence of 10 laser pulses (Table 1) and W-states with a sequence of five laser pulses (see Table 2). These pulse sequences generate three-ion entangled states within less than 1 ms. Full information on the three-ion entangled states is obtained by state tomography. For this the entangled states are subjected to 27 different sets t|S, n) —> \D,n) transition, i.e. no change of vibrational quantum number n, laser on resonance * \S, n) —• \D, n + 1), laser detuned by +u
Ill Table 1. Pulse sequence to create the GHZ-state (\GHZ) = {\DDD) + i\SSS))/V2For the definitions of Ri (9, ip) see Eq. 1 and 2. First we apply a so-called beamsplitter pulse, creating a correlation between ion # 1 and the bus mode (the phonon qubit). Ion # 2 is flipped conditional on the phonon qubit with a CNOT-operation consisting of a phase gate 3 9 enclosed in two Hadamard-like operations. Finally the phonon qubit is mapped onto ion # 3 . ion ion ion ion ion
#1 #2 #2 #2 #3
(beamsplitter) (Hadamard) (Phase gate) (Hadamard) (map)
tffCr/2,0)
«f(7T,7T/2)
J#(7r/2,0) flJ(7T,7T/2)
R+(TV/V2,0)
^'(«A") J#(ir,0)
fl+(7T,7r/2)
R+(ir/V2,0)
JtfOr.O)
Table 2. Pulse sequence to create the W-state. First we apply an asymmetric beamsplitter pulse on ion # 2 exciting the phonon mode with a probability of two-thirds. If the the phonon mode is excited, the second beamsplitter sequence removes the phonon with a probability of 0.5 and maps it onto ion # 3 . Finally, the last pulse maps the remaining phonon population onto ion # 1 and we obtain (\DDS) + \DSD) + \SDD))/V3. ion # 2 (beamsplitter) ion # 3 (beamsplitter) ion # 1 (map)
K+(2arccos(l/vr3),0) Ra{ir,ir) Ru{ir,0)
R+(n/2,n) i?+(7T,7r)
of single qubit operations before the read-out employing a CCD-camera. Prom this data all 64 entries of the density matrix are extracted with the methods described in 25>27. In total 5000 experiments —corresponding to 200 s of measurement time— are sufficient to achieve an uncertainty of less than 2% for all density-matrix elements. In Figs. 1 and 2 we show the experimental results for the density matrix elements of the GHZ- and W-states, P\GHZ) a n d P|w>- The off-diagonal elements are observed with nearly equal height as the corresponding diagonal elements and with the correct phases. Fidelities of 76% for P|GHZ) a n d 83% for /9|W) are obtained. The fidelity is defined as |(*ideai|P|exP)|*ideai)|2 with *ideai denoting the ideal quantum state and p\exp) the experimentally determined density matrix. All sources of imperfections have been investigated independently 19 and the measured fidelities are consistent with the known error budget. Note that for the W-state, coherence times greater than 200 ms were measured (exceeding the synthesis time by almost three orders of magnitude), while for the GHZ-state only ~ 1 ms was found. This is due to the W-states being a superposition of three states with the same energy. Thus, the W states are not sensitive to the overall energy scale of the system and laser
112
a)
h)
IPl Fig. 1. Real part (a), imaginary part (b) and absolute values (c) of the density-matrix elements of the experimentally obtained W-state. The off-diagonal elements are of equal height as t h e diagonal elements and indicate the coherence between the different logical eigenstates {D, S}. The fidelity is calculated to be 83 %. a) 0.5 , .
-0.5 4 DDD DDS '^ DSD x DSS
^
snr sss
DDIH
Rep
Imp
Fig. 2. Real (a) and imaginary (b) elements of a GHZ-states density matrix. The offdiagonal elements for SSS and DDD indicate the coherence. The fidelity was calculated to be 76 %.
113
frequency noise does not lead to dephasing. This is in strong contrast to the GHZ-state in Fig. 2 which is maximally sensitive to such perturbations. Similar behaviour has been observed previously with Bell-states 26 ' 27 . 4. P r o j e c t i o n of t h e q u a n t u m s t a t e s b y m e a s u r e m e n t Having tripartite entangled states available as a resource, we make use of individual ion addressing to project one of the three ions' quantum state to an energy eigenstate while preserving the coherence of the other two. Qubits are protected from being measured by transferring their quantum information into superpositions of levels which are not affected by the detection, that is a light scattering process on the Si/% —* Pi/2-transition. In Ca + , an additional Zeeman level D ' ~ D 5 / 2 (nij=-5/2) can be employed for this purpose. Thus, after the state synthesis, we apply two n pulses on the S—>D' transition of ion # 1 and # 2 , moving any S population of these ions into their respective D' level. The D and D' levels do not couple to the detection light at 397 nm (Fig. 3).
ion #3
ion #2
ion #1
Fig. 3. Partial level scheme of the three Ca-ions. Only ion # 3 is read out. Ion # 1 and #2's quantum information is protected in the Zeeman manifold of the D 5 / 2 - * eve l> namely the m j = —1/2 and mj = —5/2 levels. Note that we have labelled the ions in analogy to a binary number representation from right to left.
Therefore, ion # 3 can be read out using electron shelving 19 . After the selective readout a second set of Tr-pulses on the D' to S transition transfers the quantum information back into the original computational subspace {D, S}. For a demonstration of this method, GHZ- and W-states are generated and the qubits # 1 and # 2 are mapped onto the {D, D'} subspace. Then,
114
the state of ion # 3 is projected onto S or D by scattering photons for a few microseconds on the S-P transition. In a first series of experiments, we did not distinguish whether ion # 3 was projected into S or D. After remapping qubits # 1 and # 2 to the original subspace {S, D} r the tomography procedure is applied to obtain the full density matrix of the resulting threeion state. As shown in Fig. 4c, the GHZ-state is completely destroyed, i.e.
c)
d)
Fig. 4. Absolute values of density-matrix elements after measuring ion # 3 . (a) shows those of a GHZ-state before measuring and (c) after ion # 3 is measured. T h e same for a W - s t a t e ((b) and (d)).
it is projected into a mixture of 1555} and \DDD). In contrast, for the W-state, the quantum register remains partially entangled as coherences between ion # 1 and # 2 persist (Fig. 4c). Note that related experiments have been carried out with mixed states in NMR 14 and with photons 12 . 5. Selective read-out of a quantum register In a second series of experiments with W-states, we deliberately determine the third ion's quantum state prior to tomography: The ion string is now
115
illuminated for 500 /xs with light at 397 nm and its fluorescence is collected with a photomultiplier tube (Fig. 5a). Then, the state of ion # 3 is known
IOQO
c a
-
threshold
600
Ion#3inlS>
-
1 i
| 5£I
1
'1
[in i'.'. 1 m l D >
200
| 10
20
30
40
50
60
Photon counts in 500 s
0.4 0.5 .
0 . on
0.2 o
"N,
••*
s.- •
Dl: ' < N . lis
•' .. ••II
Fig. 5. (a) Histogram of photon counts within 500 s for ion # 3 and threshold setting. (b) and (c) Density matrix of ion # 1 and # 2 conditioned upon the previously determined quantum state of ion # 3 . The absolute values of the reduced density-matrix elements are plotted for ion # 3 measured in the S state (b) and ion # 3 measured in the D state (c). Off-diagonal elements in (b) show the remaining coherences.
and subsequently we apply the tomographic procedure to ion # 1 and # 2 after remapping them to their {S, D} subspace. Depending on the state of ion # 3 , we observe the two density matrices presented in Fig. 5b and 5c. Whenever ion # 3 was measured in D, ion # 1 and # 2 were found in a Bell state (\SD) + \DS))/V2, with a fidelity of 82%. If qubit # 3 was observed in S, the resulting state is \DD) with fidelity of 90%. This is a characteristic signature of W = (\DDS) + \DSD) + \SDD))/V3: In 1/3 of the cases, the measurement projects qubit # 3 into the S state, and consequently the other two qubits are projected into D. With a probability of 2/3 however, the measurement shows qubit # 3 in D, and the remaining quantum register is found in a Bell state 21 . Experimentally, we observe ion # 3 in D in 65 ±2%
116
of the cases. 6. Conditioned single qubit operations In section 4 we found that measuring a single qubit destroys the quantum nature of a GHZ-state completely. However, if prior to this the qubit is rotated into a different basis, the quantum nature of the GHZ-state can be partially preserved. Moreover, we can deterministically transform tripartite entanglement into bipartite entanglement using only local measurements and one-qubit operations. To demonstrate this, we first generate the GHZstate (\DSD) + \SDS))/\/2. In a second step, we apply a n/2 pulse to ion # 3 , with phase 3TT/2, rotating a state \S) to (15) - \D))/y/2 and |D) to (\S) + \D))/y/2, respectively. The resulting state of the three ions is \D)(\SD) - \DS)) + \S)(\SD) + \DS))/2. A measurement of the third ion, resulting in \D) or \S), projects qubits # 1 and # 2 onto the state (\SD) — \DS))/y/2 or the state (\SD) + \DS))/\/2, respectively. The corresponding density matrix is plotted in Fig. 6a. With the information of the state of
a)
b)
Fig. 6. (a) Real part of the density-matrix elements of the system after ion # 1 of the GHZ-state (\DSD) + \SDS))/y/2 has been measured in a rotated basis, (b) Transformation of the GHZ-state (\DSD) + \SDS))/\/2 into the bipartite entangled state |5)(|DS) + \SD))/y/2 by conditional local operations. Note the different vertical scaling of (a) and (b).
ion # 3 available, we can now transform this mixed state into the pure state \S){\SD) + \DS))/y/2 by local operations only. Provided that ion # 3 is found in \D), we perform a so-called Z-gate (Rc(ir,ir/2)RC(IT,0)) on ion # 2 to obtain \D){\SD) + \DS))/y/2 . In addition, we flip the state of ion # 3 to reset it to \S). Figure 6b shows that the bipartite entangled state \S){\SD) + \DS))/y/2 is produced with fidelity of 75%. This procedure can also be regarded as an implementation of a three-spin quantum eraser as proposed in 16 .
117
Our results show that the selective read-out of a qubit in the quantum register indeed leaves all other qubits in the register untouched. In particular that means that for certain states entanglement can be preserved in the remaining part of the quantum register. In addition, even after such a measurement has taken place, single qubit rotations can be performed with high fidelity. Such techniques mark a first step towards the one-way quantum computer 28 . The implementation of unitary transformations conditioned on measurement results has great impact as it provides a way to implement active quantum-error-correction algorithms. In addition, we will show in the next sections that it allows for the realization of deterministic quantum teleportation. 7. Teleportation Quantum teleportation exploits some of the most fascinating features of quantum mechanics, in particular entanglement, shedding new light on the essence of quantum information. It is possible to transfer the quantum information contained in a two-level system —a qubit— by communicating two classical bits and using entanglement. Thus quantum information can be broken down into a purely classical part and a quantum part. Furthermore, teleportation is not merely a simple swapping of quantum states: it does not need a quantum channel to be open at the time the transfer is carried out. Instead it uses the non-local properties of quantum mechanics (entanglement), established by a quantum channel prior to the generation of the state to be teleported. Once that link has been established, an unknown state can be transferred at any later time using classical communication only. This is quite surprising since the quantum part of the transfer seems to have happenend before the state to be transferred exists. In addition to the motivation to demonstrate and to understand quantum physics, teleportation might also have considerable impact on a future quantum computer as it facilitates the scalability of many quantum computer designs 29 . Teleportation was already demonstrated with photonic qubits 30 ' 31 ' 11 ' 32 ' 33 . However, these experiments did not include complete two-photon Bell state measurements. In addition, successful teleportation events were established by selecting the data after completion of the experiment, searching for the subset of experiments in which the outcome of the measurement and a preset reconstruction operation were matched: i.e. teleportation was performed post-selectively. In contrast to this the experiment by Furusawa et al.3i demonstrated unconditional teleportation of
118
continuous variables. Similarly Nielsen et al.35 implemented a deterministic teleportation algorithm with highly mixed states in a liquid-state NMR set-up. Recently two groups realized quantum teleportation of atomic qubits. The Boulder group 36 teleported the quantum information contained in one Beryllium-ion to another one, while the Innsbruck group 17 used Calcium ions for the same purpose. Both groups trap their ions in Paul trap, however, pursue different approaches: in Boulder the qubits are encoded in the hyperfme structure of the ions, while in Innsbruck the qubit states are stored in superpositions of a ground and metastable electronic state. Furthermore the Boulder group uses segmented traps to perform the required selective read-out of the quantum register, whereas in Innsbruck tightly focused laser beams together with selective excitation of the Zeeman levels are employed for this purpose. Finally the Boulder group chose to work with a geometric phase gate 37 , while the Innsbruck group uses composite pulses to realize the phase gate 18 . Despite these different approaches both experiments yield similar results. This demonstrates that ion traps are versatile devices for coherent state manipulation and quantum information processing. The teleportation of a state from a source qubit to a target qubit requires three qubits: the sender's source qubit and an ancillary qubit that is maximally entangled with the receiver's target qubit providing the strong quantum correlation. The quantum teleportation circuit is displayed in Fig. 7. The circuit is
u. w z H t/2 r • / / -
M>
•/A- Hide
•a
H3= JIQH Z
lEEFS
Fig. 7. The teleportation algorithm's quantum circuit. Double lines represent flow of classical information, whereas single lines flow of quantum information. The shaded lines indicate when a qubit is protected from detection light via so-called hiding-pulses. First ions # 2 and # 3 are entangled, creating the quantum link between the source region (ions # 1 and # 2 ) and the target ion (ion # 3 ) . Then after some waiting time the state to be teleported (on ion # 1 ) is prepared via the unitary operation U\- A controlled Z-gate together with detection via a photomultiplier tube (PMT) realizes the Bell state measurement. After the reconstruction pulses the success of the teleportation is tested by appling the inverse preparation procedure before measuring the target ion on an intensified CCD-camera (Charge Coupled Device).
119
formally equivalent to the one proposed by Bennett et al. 6 , but adapted to the ion-based quantum processor. It can be broken up into the following tasks: (1) Creation of Bell states A pulse sequence of three laser pulses (cf. Table 3) produces the Bellstate (\DS) + \SD)/y/2. Tomography 38 - 27 of this state shows a fidelity of up to 96% for the entangling operation. Similary to the W-states above this Bell state constists of a superpositions of states with the same energy. Indeed, we observe that the lifetime of this Bell state approaches the fundamental limit given by the spontaneous decay rate of the metastable D 5 / 2 -level of 1.2 s 27 . Now, after the quantum link between the source and the target region is established, we prepare a test state \ vla a single qubit operation U\ on the source ion. (2) Rotation into the Bell-basis The Bell-state measurement is accomplished by rotating the basis of the source and the ancilla ions into the Bell basis before the actual readout of the qubits. This rotation is implemented with a controlled-Z (phase) gate and appropriate single qubit operations. The experimental implementation of the controlled-Z-gate is described in ref. 18 . To illustrate the rotation into the Bell-basis more easily, we will use in the following a zero-controlled-not (0-CNOT) gate as a substitute for the controlled Z-gate: suppose one has the Bell state (\DS) + \SD))/y/2 (note that we use the convention \D) = |0) and \S) = |1)), then application of a 0-CNOT followed by a 7r/2-Carrier-Pulse on the control bit (the leftmost bit) yields: (|£>5> + \SD))/V2
°-CNOT) QDD) + \SD))/V2
= {\D) + \S))\D)/V2^^l\SD)
(3) (4)
The pulse i?3(7r/2,0) denotes a single qubit rotation of the control bit. Now we have mapped the Bell state \DS) + \SD) to \SD). Similarly all other Bell states are mapped onto orthogonal logical eigenstates. Therefore a measurement in the logical eigenbasis yields now a precise knowledge of the original Bell state. (3) Selective read—out of the quantum register and conditional quantum gates The measurement process must preserve the coherence of the target qubit, ion # 3 . Thus, the state of ion # 3 is hidden by transferring it to a superposition of levels which are not affected by the detection
120
light. We employ an additional Zeeman level of the D 5 / 2 manifold for this purpose. Applying now laser light at 397 nm for 250 /is to the ion crystal, only the ion in question can fluoresce, and that only if it is the Si/2~state 15 . This hiding technique is also used to sequentially read out ion # 1 and ion # 2 with a photomultiplier tube (see Fig. 3). Instead of using a CCD-camera (which can easily distinguish between different ions), we prefer to take advantage of the fast electronic readout capabilities of a photo-multiplier tube. This ensures a reaction on the measurement result within the single qubit coherence time. A digital electronic circuit counts the number of detected photons and compares it to the threshold (less than 6 detected photons indicate that the ion is in the D5/2 level). Conditioned on the measurement result, we apply single qubit rotations on the target ion 15 . This is implemented by using a classical ANDgate between the output of the electronic circuit which has stored the measurement result and the output of a digital board on which the reconstruction pulses are programmed. Thus, we apply the appropriate unitary qubit rotation, — iay, —iaz, icrx, or 1 (with Pauli operators a^) to reconstruct the state in the target ion # 3 , obtaining \ on ion # 3 . Note that — iaz is realized by applying ay followed by ax. This has the advantage that we can apply ax if ion # 1 is measured to be in \D) and az if ion # 2 is measured to be in \D). Thus the logic of the controll electronic remains quite simple. The whole pulse sequence is displayed in Table 3. In contrast to Fig.7, here also spin echo pulses are included. The conditioned pulses #31,32,33 are applied only if less than 6 photon detection events were recorded during the respective detection time of 250 /xs. The phase
121 Table 3. To implement the teleportation, we use pulses on carrier transitions R.f (6,
" §
1 .2 | i & ° .3
3
1 2 3 4 5
6
7 8 9 10 11 12 13 14 15 16 17 18
1
19
™
20 21 22 23 24 25
g
i
<2
26
27
28 29 30 , o § |
31 32
£3
33 34 35
Action Light at 397 nm Light at 729 nm Light at 397 nm Rf (7r/2,37r/2) R§(TT,3TT/2)
fl+(7T,7r/2) Wait for l^s - 10 000 /is fl?(ir,0)
^(tfx^x) fit(7r,3ir/2) fi+(7r/V^,7r/2) fl+(7T,0) fl+(7r/v/2,7r/2) Rt(ir,0) flH(7T,7r) Rj(7T,7T/2) .R^TT.O)
.R+(7r,7r/2) Hf(7r/2,37r/2) fl°(ir/2,ir/2) fl?(ir,0) PMDetection for 250 (is flH(7T,7T) PMDetection for 250 us
rt?0r,0) Wait 300 ^ s fl?(ir,7r) ^ ( 7 r / 2 , 3 7 r / 2 + ^>) R%(ir,4>) R%(n,n/2 + cl>) R§(n,4) R$Wx,
Comment Doppler preparation Sideband cooling Optical pumping Entangle ion # 3 with motional qubit Prepare ion # 2 for entanglement Entangle ion # 2 with ion # 3 Stand-by for teleportation Hide target ion Prepare source ion # 1 in state x Get motional qubit from ion # 2 Composite pulse for phasegate Composite pulse for phasegate Composite pulse for phasegate Composite pulse for phasegate Spin echo on ion # 1 Unhide ion # 3 for spin echo Spin echo on ion # 3 Hide ion # 3 again Write motional qubit back to ion # 2 Part of rotation into Bell-basis Finalize rotation into Bell basis Hide ion # 2 Read out ion # 1 with photomultiplier Hide ion # 1 Unhide ion # 2 Read out ion # 2 with photomultiplier Hide ion # 2 Let system rephase; part of spin echo Unhide ion # 3 Change basis i
this classical b o u n d a r y holds only if no assumptions on t h e states t o be teleported are m a d e . If one restricts oneself t o only t h e four test states, strategies exist which use no entanglement and yield fidelities of 78%
40
.
122
"2 fa
S) + \D) \S)+i\D) Fig. 8. Result of the teleportation: The four test states are teleported with fidelities of 76%, 74%, 73%, and 75%, respectively (bars labelled 'reconstruction'). For each input state 300 single teleportation experiments were performed. The error of each entry, estimated from quantum projection noise, is 2.5%. The bars labelled 'no reconstruction' show the results if the reconstruction operations are omitted, yielding an average fidelity of 49.6%. The optimum teleportation obtainable by purely classical means and no assumptions about the initial states reaches a fidelity of 66.7% (dashed line).
However, each of these strategies must be designed for a specific test state ensemble to work properly. Note also that, in order to rule out out hidden variable theories, a fidelity in excess of 0.87 is required 42 . For comparison, we also show data where the reconstruction pulses were not applied. Without the classical information obtained from the Bell state measurement, the receiver's state is maximally mixed, i.e. there is no information available on the source state. Also, the measurement outcome of ions # 1 and # 2 does not contain any information about the initial state. Indeed we find each possible result with an equal probability of 0.25±0.036, independent of the test input states. Note that only with both the receiver's qubit and the result of the Bell measurement, the initial state can be retrieved. We emphasize that the conditional, deterministic reconstruction step, in combination with the complete Bell state analysis, is one of the crucial
123
improvements with respect to former experimental realizations of quantum teleportation. Furthermore, after the teleportation procedure the state \ 1S always available and may be used for further experiments. To emphasize the role of the shared entangled pair as a resource, we store the Bell state for some time and then use it only later (after up to 20 ms) for teleportation. For waiting times of up to 20 ms (exceeding the time we require for the teleportation by a factor of 10) we observe no decrease in the fidelity. For longer waiting times, we expect the measured heating of the ion crystal of smaller than 1 phonon/100 ms to reduce the fidelity significantly. This is because for a successful rotation into the Bell-basis we require the phonon number in center-of-mass mode of the ion string to be in the ground state. The obtainable fidelity is limited mainly by dephasing mechanisms. The most obvious one is frequency fluctuations of the laser driving the qubit transition, and magnetic field fluctuations. Since these fluctuations are slow compared to the execution time of 2 ms, they can be cancelled to some degree with spin echo techniques 43 . However, during the algorithm we have to use different pairs of states to encode the quantum information, one of which being only sensitive to magnetic field fluctuations while the other one being sensitive to both laser and magnetic field fluctuations. To overcome these complications, two spin echo pulses are introduced (see Table 3). Their optimal position in time was determined with numerical simulations. From measurements we estimate that the remaining high frequency noise reduces the fidelity by about 5%. Another source of fidelity loss is an imperfect AC-Stark shift compensation. AC-Stark compensation is needed to get rid of the phase shifts introduced by the laser driving the weak sideband transition due to the presence of the strong carrier transitions 44 . Recent measurements suggest that an imperfect compensation as introduced by the incorrect determination of the sideband frequency by only 100 Hz lead to a loss of teleportation fidelity on the order of 5%. Imperfect state detection as introduced by a sub-optimal choice for the threshold (6 instead of 3 counts) was analyzed later to contribute on the order of 3% to the fidelity loss. However, the biggest contribution to the read-out error stems from an incorrect setting of the hiding pulse frequency and strength. It reduced the fidelity by 7%. Addressing errors on the order of 3-4% were estimated via numerical simulations to reduce the fidelity by about 6%. The addressing errors were measured by comparing the Rabi flopping frequency between neighboring ions and correspond to a ratio of 10~ 3 in intensity between the addressed
124
ion and the other ones. Treating these estimated error sources independently (multiplying the success probabilities) yields an expected fidelity of 77% in good agreement with the experimental findings. In conclusion, we described an experiment demonstrating teleportation of atomic states. The experimental procedures might be applied in future quantum information processing networks: with long lived entangled states as a resource, quantum teleportation can be used for the distribution of quantum information between different nodes of the network. We gratefully acknowledge support by the European Commission (QUEST and QGATES networks), by the ARO, by the Austrian Science Fund (FWF). H.H. acknowledges funding by the Marie-Curie-program of the European Union. T.K. acknowledges funding by the Lise-Meitner program of the FWF. We are also grateful for discussions with A. Steinberg and H. Briegel.
References 1. M. A. Nielsen, I. L. Chuang, Quantum Computation and Quantum Information (Cambridge Univ. Press, Cambridge, 2000). 2. P. W. Shor, Phys. Rev. A 52, R2493 (1995). 3. A. M. Steane, Phys. Rev. Lett. 77, 793 (1996). 4. C. H. Bennett, D. P. DiVincenzo, J.A. Smolin, W. K. Wootters, Phys. Rev. A 54, 3824 (1996). 5. A. M. Steane, Nature 399, 124 (1999). 6. Bennett, C. H., G. Brassard, C. Crepeau, R. Jozsa, A. Peres, and W. K. Wootters, Phys. Rev. Lett. 70, 1895 (1993). 7. J. Chiaverini, D. Leibfried, T. Schaetz, M.D. Barrett, R.B. Blakestad, J. Britton, W.M. Itano, J.D. Jost, E. Knill, C. Langer, R. Ozeri & D.J. Wineland, Nature 432, 602 (2004). 8. A. Rauschenbeutel, G. Nogues; S. Osnaghi, P. Bertet, M. Brune, J.-M. Raimond, S. Haroche, Science 288, 2024 (2000). 9. C.A. Sackett, D. Kielpinski, B.E. King, C. Langer, V. Meyer, C.J. Myatt, M. Rowe, Q.A. Turchette, W.M. Itano, D.J. Wineland, and C. Monroe, Nature 404, 256 (2000). 10. J. W. Pan, D. Bouwmeester, M.Daniell, H. Weinfurter, A. Zeilinger, Nature 403, 515 (2000). 11. J. W. Pan, M. Daniell, S. Gasparoni, G. Weihs, A. Zeilinger, Phys. Rev. Lett. 86, 4435 (2001). 12. M. Eibl, N. Kiesel, M. Bourennane, C. Kurtsiefer, H. Weinfurter, Phys. Rev. Lett. 92, 077901 (2004). 13. Z. Zhao, Y.-A. Chen, A.-N. Zhang, T. Yang, H. Briegel, J.-W. Pan, Nature 430, 54 (2004).
125
14. G. Teklemariam, E. M. Fortunato, M. A. Pravia, Y. Sharf, T. F. Havel, D. G. Cory, A. Bhattaharyya, and J. Hou , Phys. Rev. A 66, 012309 (2002). 15. C. F. Roos, M. Riebe, H. Haffner, W. Hansel, J. Benhelm, G. P. T. Lancaster, C. Becher, F. Schmidt-Kaler, R. Blatt, Science 304, 1478 (2004). 16. R. Garisto and L. Hardy, Phys. Rev. A 60, 827, (1999). 17. M. Riebe, H. Haffner, C. F. Roos, W. Hansel, J. Benhelm, G. P. T. Lancaster, T. W. Korber, Nature 429, 734 (2004). 18. F. Schmidt-Kaler, H. Haffner, M. Riebe, S. Guide, G. P. T. Lancaster, T. Deuschle, C. Becher, C. F. Roos, J. Eschner & R. Blatt, Nature 422, 408 (2003). 19. F. Schmidt-Kaler, H. Haffner, S. Guide, M. Riebe, G. P.T. Lancaster, T. Deuschle, C. Becher, W. Hansel, J. Eschner, C. F. Roos, R. Blatt, Appl. Phys. B: Lasers and Optics 77, 789 (2003). 20. D. M. Greenberger, M. Home, A. Zeilinger, in Bells Theorem, Quantum Theory, and Conceptions of the Universe, edited by M. Kafatos (Kluwer Academic, Dordrecht, 1989). 21. W. Diir, G. Vidal, J. I. Cirac, Phys. Rev. A 62, 062314 (2000). 22. A. Zeilinger, M. A. Home D. M. Greenberger, NASA Conf. Publ. No. 3135 National Aeronautics and Space Administration, Code NTT, Washington, DC, (1997). 23. J.I. Cirac and P. Zoller, Phys. Rev. Lett. 74, 4091 (1995) 24. M. Sasura and V. Buzek, J. Mod. Opt. 49, 1593 (2002). 25. D. F. V. James, P. G. Kwiat, W. J. Munro, A. G. White, Phys. Rev. A 64, 052312 (2001). 26. D. Kielpinski, V. Meyer, M.A. Rowe, C.A. Sackett, W.M. Itano, C. Monroe, D.J. Wineland, Science 291, 1013 (2001). 27. C. F. Roos, G. P. T. Lancaster, M. Riebe, H. Haffner, W. Hansel, S. Guide, C. Becher, J. Eschner, F. Schmidt-Kaler, R. Blatt, Phys. Rev. Lett. 92, 220402 (2004). 28. R. Raussendorf and H. J. Briegel, Phys. Rev. Lett. 86, 5188 (2001). 29. D. Gottesman D. & I.L. Chuang, Nature 402, 390 (1999). 30. D. Bouwmeester, J-W. Pan, K. Mattle, M. Eible, H. Weinfurter, and A. Zeilinger , Nature 390, 575 (1997). 31. D. Boschi, S. Branca, F. DeMartini, L. Hardy and S. Popescu , Phys. Rev. Lett. 80, 1121 (1998). 32. I. Marcikic, H. de Riedmatten, W. Tittel, H. Zbinden, & N. Gisin, Nature 421, 509-513 (2003). 33. D. Fattal, E. Diamanti, K. Inoue and Y. Yamamoto, Phys. Rev. Lett. 92, 037904 (2004). 34. A. Furusawa, J.L. S0rensen, S.L. Braunstein, C.A. Fuchs, H.J. Kimble, E.S. Polzik, Science 282, 706 (1998). 35. M. A. Nielsen, E. Knill & R. Laflamme, Nature 396, 52 (1998). 36. M. D. Barrett, J. Chiaverini, T. Schaetz, J. Britton, W.M. Itano, J.D. Jost, E. Knill, C. Langer, D. Leibfried, R. Ozeri & D.J. Wineland, Nature 429, 737 (2004). 37. D. Leibfried , B. DeMarco, V. Meyer, D. Lucas, M. Barrett, J. Britton, W.
126
38. 39. 40. 41. 42. 43. 44.
M. Itano, B. Jelenkovic, C. Langer, T. Rosenband & D. J. Wineland, Nature 422, 412 (2003). K. Vogel and H. Risken, Phys. Rev. A 40, 2847 (1989). S. Guide, M. Riebe, G.P.T. Lancaster, C. Becher, J. Eschner, H. Haffner, F. Schmidt-Kaler, I.L. Chuang & R. Blatt, Nature 421, 48 (2003). Steven van Enk, private communication. S. Massar and S. Popescu, Phys. Rev. Lett. 74, 1259 (1995). N. Gisin, Physics Letters A 210, 157 (1996). E.L. Hahn, Phys. Rev. 77, 746 (1950) H. Haffner, S. Guide, M. Riebe, G. Lancaster, C. Becher, J. Eschner, F. Schmidt-Kaler, and R. Blatt, Phys. Rev. Lett. 90, 143602-1-4 (2003).
127
Liquid-State N M R Quantum Computer: Hamiltonian Formalism and Experiments Yasushi KONDO* and Mikio NAKAHARA Department of Physics, Kinki University Higashi-Osaka 577-8502, Japan * E-mail: [email protected] Shogo TANIMURA Graduate School of Engineering, Osaka City Osaka 558-8585, Japan
University
A quantum algorithm is represented as a unitary operator acting on a qubit space and is realized as a time development operator of a controlled quantum system. In a liquid-state NMR quantum computer, spins of nuclei in isolated molecules dissolved in a solvent work as qubits and they are controlled by radio frequency pulses. We reexamine the Hamiltonian describing the spin dynamics and formulate the principle of a liquid-state NMR quantum computer from the viewpoint of Hamiltonian dynamics. We executed the Deutsch-Jozsa algorithm and the Grover database search algorithm as well as a pseudopure state preparation with two-qubit homonucleus molecules to demonstrate the validity of our formalism. Finally we discuss whether a liquid-state NMR quantum computer currently fulfills the DiVincenzo criteria or not.
1. Introduction Quantum computation has been recently attracting a lot of attention since it is expected to solve some of computationally hard problems for a conventional digital computer. 1 Various physical systems for its realization have been proposed to date. Above all, a liquid-state NMR (nuclear magnetic resonance) is considered to be most successful owing to a beautiful demonstration of Shor's factorization algorithm. 2 Disadvantages of a liquid-state NMR quantum computer have been also pointed out. 3 For example, the number of qubits is possibly limited up to about ten because of poor spin polarization at room temperature and overlapping chemical shifts. In spite of these limitations, the liquid-state NMR quantum computer is the best test bench to study implementation of quantum algorithms at the moment.
128
Therefore it is important to understand its dynamics from a Hamiltonian viewpoint. Suppose we would like to implement a quantum algorithm whose unitary matrix representation is C/aig. Our task is to find the control parameters j(t) in the Hamiltonian H such that the time development operator U[1(t)}=Texp
•j.j
H(-y(t))dt
(1)
is equal to £/aig, where T stands for the time-ordered product. This contribution is organized as follows. In Section 2 we explain the principle of NMR measurements. We put emphasis on the role of a Hamiltonian that governs spin dynamics. In Section 3 control of two-qubit molecules is discussed. Since two qubits are minimal for performing nontrivial quantum computation, two-qubit systems are worth discussing in detail. Issues related to the density matrix are discussed in Section 4. In Section 5 quantum computations, by employing cytosine as an example, are discussed. Section 6 is devoted to discussion on the DiVincenzo criteria? and summary. First four sections lay foundation of NMR quantum computing while Sections 5 and 6 deal with more advanced topics. Other aspects of liquidstate NMR quantum computing are available in excellent reviews. 5-8 2. N M R Primer Here we describe a typical experimental setup for NMR measurement and explain its theoretical basis. We exclusively analyze molecules with a single spin-1 nucleus in this section while molecules with more spins will be studied in the following sections. i
2.1. Experimental
setup
NMR is an experimental method to measure and control a state of nuclear spins in a molecule. Molecules in the sample tube are moving and rotating rapidly and randomly and hence the intermolecular spin-spin interactions are averaged out to vanish while intramolecular spin-spin interactions must be taken into account when writing down the Hamiltonian. Suppose a static magnetic field is applied on a sample that contains nuclear spins. Then the eigenstates of the spins are defined with respect to the magnetic field. If the spins are not in one of the eigenstates, they exhibit a precession, which is observed as a rotating magnetization of the sample. When an oscillating magnetic field with frequency equal to the energy difference in two spin
129
test tube
Fig. 1. NMR setup. Molecules in the test tube are moving and rotating rapidly and randomly.
states is applied in addition, the spin state continuously changes. The final state of spins is also measured as a rotating magnetization. An experimental setup for NMR measurement, shown schematically in Fig. 1, consists of three parts. The first part contains a superconducting magnet system to provide a homogeneous constant magnetic field and a field gradient coil system to produce temporally controlled field gradients. The second is a resonance circuit to apply rf (radio frequency) magnetic fields on the sample and pick up rf signals from rotating magnetization. The third is an assembly of electronic circuits to produce rf pulses and detect rf signals. 9 ' 10 The electric oscillator generates a continuous alternating current with the radio frequency wrf and the sequencer modulates the oscillating current to generate rf pulses. A typical temporal duration of a pulse, which is called a pulse width, is of the order of 10 fis. The rf pulses are amplified and fed into the resonance circuit, which generates rf magnetic fields on the sample in the test tube. The directional coupler is inserted to prevent direct transmission of the rf pulses into the receiver. The rotating magnetization of the sample induces signals at the coil, which are lead to the receiver and detected. 2.2. Hamiltonian
for a single
spin
There are a large number, typically ~ 10 20 , of molecules in the test tube. Sample molecules interact with each other and also with the solvent molecules. These interactions cause random motion of the molecules. However, these motions are very rapid, of the order of 1 ps, and thus spin-spin
130
interactions among these molecules are temporally averaged to vanish. In other words, we may assume that spin degrees of freedom of each molecule are effectively isolated from each other. Therefore, we only need to consider a single molecule Hamiltonian and totally ignore the interactions among molecules. This is a great merit of a liquid-state NMR. Let us consider the Hamiltonian H of a single spin in a static magnetic field along the z-direction and the rf magnetic field along the x-direction, see Fig. 1. The Hamiltonian H is decomposed into two terms as H = Ho + HT{ H0 = -Hu>0Iz,
HTf = 2hu>i cos(wrf£ -
(2)
where Jj = (Ji/2 and <7j is the Pauli matrix indexed by i = 1, 2, 3 = x, y, z. The parameter wo is proportional to the strength of the magnetic field Bo and is given by u>o = ~fBo with the gyromagnetic ratio 7 of the nucleus. The first term Ho describes action of the static magnetic field on the spin. In the static magnetic field the spin precesses with the Larmor frequency woThe second term HTf describes the action of the rf pulses on the spin and hence it is controllable. Control parameters are the time-dependent field strength w\, the frequency wrf, and the phase
1
i=0 j = 0
where |0) and |1) denote the eigenstate of the spin j (lower energy state) and J, (higher energy state), respectively. The ensemble average of an observable M is given by 1
tr(pM)= X > « 0 ' | M | t > .
(4)
i,j=0
The density matrix pth of the thermal state without rf magnetic fields is (see, Eq. (2)) e-(H0/kBT)
Pth
=
tr [e-(tfo/fcBT)j
e-(-hu>0Iz/kBT)
=
tr [ e -(-^o/,/fc B T)j (5)
131
The above approximation may be justified since huQ/ksT ~ 10~ 5 at room temperature. The time development of p(t) is determined by the Liouville equation ih^
= [H,p} = Hp-PH.
(6)
We introduce a time-dependent unitary operator U(t) and define the unitary transformation of the density matrix as p = UpUl
(7)
ihft=[H,p\
(8)
Then Eq. (6) becomes
with the transformed Hamiltonian H = UHU1 -ihU^-.
(9) at
If we take the unitary transformation U = ex.p(-idjIotIzt),
(10)
then p represents the spin state viewed in a frame rotating with the frequency uirot around the z-axis. The transformed Hamiltonian becomes H = H0 + Hlf with H0 = -H(UJ0 -
ujTOt)Iz
and HTf = hu>i [{cos
Q
J .
The above Hamiltonian is simplified if we take u>Iot = wrf as H0 = -h{uj0 - u)l{)Iz Hr{ = fkoi [{cos
2uTft)}Iy].
The typical time scale 1/wi of the spin dynamics is much longer than the period of the oscillating field such that 1/wi 3> l/u>rf. Thus the terms rapidly oscillating with frequency 2wrf in Hr{ are averaged to vanish. Therefore, HT{ is approximated as Hr{ = ftw\ (cos<j)Ix +
singly).
132
in the rotating frame. Moreover, when the resonance condition U>Q = wrf is satisfied, Ho vanishes and the total Hamiltonian reduces to H = HTf = tva)\ (cos cj)Ix + sin
(11)
Therefore, the spin dynamics becomes simpler if it is viewed in the rotating frame with wrot = wrf = U>Q. The fact that the factor 2 in the Hamiltonian HTf (Eq. (2)) disappears in the transformed Hamiltonian HT{ (Eq. (11)) is physically understood as follows.9 An oscillating rf magnetic field is considered to be a superposition of two rotating fields of ±wrf and the effect of the rotating field with —wrf is averaged out to vanish in our rotating frame. 2.3.
Dynamics
Let us apply the previous argument to analyze the spin dynamics. The density matrix pth in Eq. (5) represents the thermal state in the laboratory frame. It is transformed to pth in the rotating frame as
= exp(-iw0Izt)
l / hw0 \ - I I + -j—f l* J exp(iu)0Izt)
Actually p t h is identical to pth since both 1/2 and Iz commute with exp(±iLOotIz)- Suppose that an rf magnetic field with phase <j> = 0 is applied for the period of r so that u>\T = 7r/2. The reduced Hamiltonian in the rotating frame is H = HwJz,
(13)
which represents an effective magnetic field parallel to the x-axis. The time development operator (1) now becomes exp ( - -
/ Thwilxdt\
= exp ( - ^ 4 j •
( 14 )
The density matrix pr! obtained after the rf pulse is applied to p t h = pth then becomes
'"•<=exp
(-'i '•) \y + 0 '•)exp ('!'•)
133
Physically, the magnetization is turned from the z-direction to the — ydirection by the effective field parallel to the z-axis in the rotating frame. We call this pulse a 7r/2-pulse around the x-axis. Then, the density matrix prf in the laboratory frame after the application of this rf pulse is
= exp(iaj0Izt) - f I - -j—f h ) exp(-io;o/ z f)
2
4kBT
yie-*"*'
0
K
J '
'
We calculate the expectation value of magnetization in the ^-direction, which generates the signal from the sample. The result is proportional to trC^Prf) = - ^ - | ; sin(woi).
(17)
This represents the magnetization oscillating with a frequency UJQ. 2.4. Quantum
state
tomography
The density matrix is Hermitian and thus generally parametrized as 1 / l + c a — ib\ I +aI =2{a + i b l - c ) = 2 *
,T y +
T
P
+ bI
T Cl
>
,, oN (18)
by real parameters a,b, c ~ fkvo/2kBT. Note that 7/2 is included for normalization (trp — tr(J/2) = 1). By measuring all the parameters a, b, and c, we know the state of the system completely. The technique to measure them is called quantum state tomography. The density matrix p in the laboratory frame at time t is p(t) = U\t)~pU{t) = exp(iu>otIz)p exp(—iwotIz) If 1+c (a-ib)e^\ iwot 2 V(a + ^ ) e 1-c J '
{
'
Therefore, the z-component of the magnetization in the laboratory frame is M(t) oc tr(o-x p) = a cos tj0t + b sin w 0 i.
(20)
The density matrix after the application of a 7r/2-pulse along the z-axis in the rotating frame is e-i*ix/2p
g^/,/2
=
I
+ aIx
_
cIy
+ bIz
(21j
134 400
(a) >
400
f(o>)
(b) 300
200l
20O
100 0.96
CO
100
0.98
1.02
1.04
0.96
0.98
X-Tffi
-100
-100
-200
-200
1.04
Fig. 2. Cosine Fourier transformations /(a>) of free induction decay signals (a) / ( t ) = e~'/ T 2 cosuot and (b) / ( t ) = e ~ ' / T 2 sinwot- Here wo = 1 and 1/T 2 = 0.001.
and it follows from Eq. (20) that the x-component of the magnetization in the laboratory frame is M(t) oc acosuot — csimoot.
(22)
Similarly the density matrix after the application of a 7r/2-pulse along the j/-axis in the rotating frame is e-i*iv/2p
e^/v/a = I + dx + bly - alz
(23)
and the rc-component of the magnetization in the laboratory frame is M(t) oc ccoswo^ + b sin wot.
(24)
Therefore, a, b, and c can be measured with a phase sensitive detector, as we show in the next section. 2.5.
Measurements
There are relaxation mechanisms in reality and a signal induced in the coil lasts only for a limited period of time. The signal is called the free induction decay (FID) signal after pulses and its time constant is denoted as T2. The signal is well approximated by f(t) = e-*/ T 2 cos(uj0t +
(25)
The cosine Fourier transformation of f(t) = e~l/T2 coscuot and f(t) = e -t/T 2 s m u ; 0 £ a r e shown in Fig. 2. From FID signals generated by the oscillating magnetization (20), we find the coefficients a and b by Fourier transformation. Similarly, by applying the pulses along the x- and y-directions, we measure the other coeffi-
135
cients in Eqs. (22) and (24). The components in the signal with coswo^ and sinwo* are often called real and imaginary parts, respectively, in NMR. 2.6. Operations
in rotating
frame
2.6.1. Unitary operations Suppose that an rf magnetic field with the phase
(26)
The thermal state evolves into
prf = u^e)pthul(9) - r +
, ° [sin 9(1x sin
(27)
after the pulse is applied. This shows that the magnetization has been rotated by an angle 9 around the axis (cos (f>, sin
(28)
that rotates the spin by an angle tp around the z-axis. The generator Iz does not exist in the Hamiltonian (11) in the rotating frame and hence it seems to be impossible to rotate a spin around the z-axis by any pulse. There are, however, several ways to rotate a spin around the z-axis. One way is to apply a sequence of three pulses C7b(7r/2)t^/2(e)C/w (TT/2). The product is exp ( - * | / z ) exp (-i9Iy) exp (i^Ix)
= exp (-iOIz) = Uz(9).
(29)
Another sequence U«,2{*n)U0{-0)Uto/2{Tt/2)
= Uz(9)
(30)
also rotates a spin by 9 around the 2-axis. Another method to deal with the spin rotation around the z-axis in the rotating frame uses the relation U*{0)Ux{,l>) = Uz{i))U
(31)
136
which is easily proved. Repeated use of the above relation makes it possible to shift Uz(ip) toward the end of the pulse sequence. The rotation around the z-axis at the end (or beginning) of the pulse sequence can be reinterpreted as a phase of a quantum system, which is not observable. Therefore the rotation around the z-axis may be, in fact, dropped. Finally we have a complete set of one-qubit operations, which turn a spin to any directions. 2.6.2. Operators in actual pulse sequences In practical applications, pulses that rotate a spin by 7r/2 around the x-, —x-, y-, —y-, z- and — z-axes are commonly used. Let us summarize here the properties of these operations. The symbols X,Y,Z,X,Y and Z denote the rotations of the spin by 7r/2 around the x-, y-, z-, —x-, —y- and —z-axes, respectively. Their explicit forms are
X = U0(«/2) = ^ (_*. 7 ) , Y = CV a (*/2) = 75 ( \ ~i ) '
X = UAn/2) = ^ ( j }) , ?
= ^/a(*/2) = ^
( _ \ \ ) - (32)
The following relations are useful to simplify pulse sequences: XYX = Z, XYX = Z, XY = ZX, XZ = ZY, XZZ = ZZX,
YXY = Z, YXY = Z, XY = YZ, YZ = ZX,
XYX = Z, YXY = Z XYX = Z, YXY = Z YX = XZ, YX = ZY XZ = ZY, YZ = ZX YZZ = ZZY.
(33)
The Hadamard gate Hi, which often appears in quantum algorithms, takes the form
and is constructed as a product H = iYZZ.
(35)
Note that Hi is often approximated by Y in actual NMR pulse sequences for quantum algorithms. This approximation breaks down if the input state is a superposition of |0) and |1).
137
2.6.3. Field gradient A field gradient is the inhomogeneity in strength of the static magnetic field that is applied to the sample. Field gradients are employed in quantum computation to introduce non-unitary operations in preparation of a pseudopure state, for example. We consider single spin dynamics in the presence of a field gradient. A field gradient modifies the system Hamiltonian (2) as H0 = -hw0(l
+ Vz)Iz,
(36)
where V is the normalized strength of the field gradient along the zdirection. We assume that the sample extends over — LQ < z < LQ. In the frame rotating with frequency U/Q, the relevant Hamiltonian depends on z as HQ = -hVoJozh
(37)
and thus the time development operator (1) is UG{Z) = exp I — — IT — hVtjQzIzdt = exp (iVu>ozIzT).
(38)
After application of the field gradient for a time interval r, the density matrix J)Q{Z) in the rotating frame becomes ^-dependent as pG{z) =
UG{z)pUG{z)
= exp (iVuj0zTlz) ~ 2+Cz+\(a
(- +alx + bly + clz J exp + ib)e-™"ozr
0
(-iVuJ0ZTlz) J
(M)
for the initial density matrix p = 1/2 + alx + bly + clz. Since the signal picked up by the coil is a sum of the contributions from all the sample molecules, pG(z) should be averaged over the sample size —LQ < z < LQ. The third term in the last line of Eq. (39) is averaged to vanish if the condition LQVUJQT 3> 1 is satisfied. Therefore, application of a field gradient is equivalent to a non-unitary operation that removes off-diagonal elements of the density matrix as Da • P = ~ + alx + bly + clz .-> — - / ^
^M) J-Lo
UG{z)~pU^G{z)dz =
l
-+dz.(40)
138
2.6.4. Block-Siegert effect Bloch-Siegert effect was originally introduced to take into account the effect of a counter rotating rf field which is time-averaged to be zero in Eq. (II). 1 1 Ramsey then generalized it to the case when two or more rf fields with different frequencies are applied to the sample. 12 Although multiple rf fields are dealt in both the original and generalized Bloch-Siegert effects, its essential effect can be considered as an influence of a widely off-resonance rf field with the angular velocity u o + A on a spin with the resonance frequency 0J0.
Let us start with a Hamiltonian in the laboratory frame, HBS = -fruoh
+ 2hui cos ((w0 + A)i) Ix,
(41)
where we took wrf = wo + A and 0 = 0 in (11). To simplify our calculation, we take U = exp [-i(u0 + A)t/ Z ]. Then, we obtain the Hamiltonian in the rotating frame with the angular velocity u>0 + A, HBS = HAIz+thv1Ix,
(42)
where rapidly oscillating terms are averaged to vanish. Since (42) is timeindependent, the time-development operator Eq. (1) is calculated as
where e = wi/A. When we assume A > ui, i.e. e < 1, (43) becomes exp
f THBSdt\
=
e
--
A
^+^.
(44)
The meaning of Eq. (44) is (a) the applied off-resonance rf field does not rotate the spin and (b) the Larmor frequency is apparently shifted by A(l — \J\ + e 2 ). Although the shift of the Larmor frequency is small, | r A ( l — v 7 ! + t2)\ is not necessarily small because r can be large.
139
In NMR quantum computation with homonucleus molecules, the BlochSiegert effect sometimes have to be taken into account. Even though a soft selective pulse is easily designed so that it does not practically rotate the other spins, it may still affect the phases of the other spins, see Eq. (44). Therefore, it causes the errors in the phase of the subsequent (or, simultaneously applied) pulses on the other spins. Let us consider three examples of the Bloch-Siegert effect on a two-spin homonucleus molecule. We call the two spins A with the Larmor frequency wo A and B with WOB = <^OA + A. (1) Square 7r-pulse with the pulse width 2 • 27r/A: The pulse on B causes the approximate phase shift 1 (7T/(2 • 2 7 T / A ) ) 2
-^^-^—A 2
,„
„
/AX
7T
• (2 • 2TT/A) = —- ~ -0.39 v ' ; 8
A on A. (2) Square 7r-pulse with the pulse width 4 • 2TT/A: The pulse on B causes the approximate phase shift 1 (TT/(4 -T;^-^—A 2
• 2TT/A)) 2 , , „ . . , • v(4 • 2TT/A) = A ' ;
n ~ 16
-0.20
on A. (3) Square 7r/2-pulse with the pulse width 4 • 27r/A: The pulse on B causes the approximate phase shift
- 1 M M 2
A
-(4v .2,/A)—i K_
64
on A. The soft pulse which we employ in the following sections is a Gaussian pulse9 and its Bloch-Siegert effect is about 1.7 times larger than that of the square pulse with the same pulse width. We mainly employed the n/2Gaussian pulses with the pulse width of 4 • 27r/A, and thus we can, as a first approximation, neglect the Bloch-Siegert effect in the examples we will discuss in the following sections. More sophisticated shaped pulses such as UBURP 9 are sometimes employed by others. We note that they affect the NMR quantum computation more than Gaussian pulses do since they have much larger peak strengths of the rf field than Gaussian pulses.
140
3. Hamiltonians for two-qubit molecules We consider molecules with two spins here since any nontrivial quantum computation requires at least two qubits. There are two classes of molecules with multiple spins; heteronucleus molecules and homonucleus molecules. Heteronucleus molecules are easy to control. However the number of qubits therein is practically limited to two or three due to the limitation inherent in an NMR spectrometer.
3.1. Heteronucleus
molecules
3.1.1. Experimental Setup The NMR setup for heteronucleus molecules is shown schematically in Fig. 3. A typical example of a two-qubit heteronucleus molecule is 1 3 Clabeled chloroform, for which 13 C and H nuclei are qubits. Because the resonance frequencies u>o,i, i = 1,2 for the first and second spins, respectively, are largely different, two resonance circuits and two assemblies of electronic circuits are required. The difference in the resonance frequencies, A(j 0 = (^o,2 — w o,i), allows us to address individual spins in quantum computation. We assume, without loss of generality, that Awo > 0. The oscillator 1 (2) generates an oscillating electric current with frequency wo,i (^0,2)- The sequencer 1 (2) modulates the oscillating current to generate rf pulses. A typical temporal duration of a pulse, called a pulse width, is on the order of 10 fis. The rf pulses are amplified and fed into the resonance circuit 1 (2), which generates rf magnetic fields on the sample in the test tube. The directional couplers prevent the rf pulses from being transferred to the receiver 1 (2). The signals, which are induced at the coils by the rotating magnetizations of the sample, are led to the receiver 1 (2) and detected.
3.1.2. Hamiltonian in rotating frame The two-qubit Hamiltonian in the laboratory frame is decomposed into three parts as H = #o + ffrf,l + #rf,2.
(45)
The first term Ho describes dynamics of the spins in the static magnetic field and is called the system Hamiltonian after the reference.7 The rest -^rf.i (i = 1,2) is a Hamiltonian controllable by rf pulses.
141
coil 2
directional coupler [Receiver^
Fig. 3.
NMR setup for heteronucleus molecules.
The system Hamiltonian Ho in the laboratory frame is Ho = - M u {Iz ®I)~
^aa
{I <8> Iz) + HJ J2
& ® 7 »)'
(46)
where I is the unit matrix of dimension two. The first two terms generate free precession of spins while the third term represents the interaction between the two spins with the coupling strength J. We will not discuss the origin of this interaction. The reader should refer to standard NMR textbooks 9 ' 1 0 for further details. The controllable Hamiltonians HTfi are HT{ i = 2fuvi i cos(wrf it — fa) I Ix <E> / H —/ ® Ix wo,i
(47)
ffrf)2 = 2fiwii2 cos(o;rf,2i - fa) ( ——Ix ® I + I ® Ix ^0,2
where o>o,2/^0,1 is the ratio of the gyromagnetic ratios of two spins. The amplitudes w\ti, the frequencies Wrf^, and the phases fa of the rf fields are control parameters. Note that the rf magnetic field is oscillating along the x-axis in the laboratory frame, see Fig. 3.
142
Let us analyze the Hamiltonian in a rotating frame next. The unitary operator U in Eq. (10) for transformation to a rotating frame is replaced with U(t) = e~iu-"'t-iI't
® e - iw ">t,2/zt
(48)
if there are two spins. We may choose different values for wrot,i and wrot,2The transformed density matrix p = UpW represents the state of the system viewed from a frame rotating with frequency 0JIot,i for the spin i. The Hamiltonian in the rotating frame becomes H = UHUX
-ihU^-U^ dt = Ho + Hritl+HvU2.
(49)
The transformed system Hamiltonian is H0 = UHoU1 -
ihU^-U] dt = -fi(w 0 ,l - W rot ,l) (Iz <8> I) - fi(w0,2 - W rot ,2) (I ® Iz)
/0 +hJ(lz®lz)
hj
0
2
0 0
Q gi(Wrot,l—wrot,2)t
Vo
o
0
0\
g —*(^rot,l—"rot^)* Q Q
Q
o
0/ (50)
It is convenient to set the rotation frequencies o>rot,i equal to the Larmor frequencies wo,»- Then the first two terms in Eq. (50) vanish. Let ALUQ be the difference between Larmor frequencies; Awo =a>o,2-wo,i-
(51)
The condition Au>o 3> J is always satisfied for heteronucleus molecules, and hence the last term vanishes if it is averaged over the time scale T such that AUQ1 < T < J " 1 . For example, Au0 ~ 400 MHz, while J ~ 200 Hz for 13 C-labeled chloroform at 11 T. Therefore, after setting u>Tot,i = w o,i, HQ is approximated as H0 = HJ {Iz ® Iz).
(52)
Let us consider the controllable Hamiltonians next. They are simplified if we set w roti j = Wo,i = wrf,» &s #rf,l = fell [COS (f>i (Ix ® I) + Sin (/>i (Iv (g> I)] , (53) #rf,2 = ^ 1 , 2 [COS (/)2 (I
143
after eliminating terms rapidly oscillating with the frequencies 2u>o,i or AUJQ. It is important to note that pulses with the frequency uTfti influence only the spin i but has no effect on the other spin in the rotating frame. This is because the difference Aa>o in the resonance frequencies is much larger than the inverse of the typical pulse width ~ 1/(10 • 10" 6 ) s" 1 ~ 100 kHz. In other words, the pulses with frequency u>rfti does not contain the Fourier component which resonates with the other spin. See also § 2.6.4. In conclusion, the Hamiltonian for heteronucleus molecules in the rotating frame is H =
hJ(Iz®Iz) +hcj\ii [cos >i (Ix
(54)
when the condition wrot,» = u>o,i = Wrf^ is satisfied. 3.2. Homonucleus
molecules
3.2.1. Experimental Setup A typical NMR setup with homonucleus molecules is shown schematically in Fig. 4. Cytosine solved in D2O is employed as a typical example of a sample molecule, where two H nucleus spins are the qubits. When |Aw 0 | is small compared to w0,i, a common resonance circuit and a power amplifier may be employed to control both spins. For example, AOJQ ~ 800 Hz for cytosine in D 2 0 when wo,i = 500 MHz. However, this small difference in wo,i still allows us to address each spin individually. The oscillator 1 (2) generates an electric current oscillating with the radio frequency wo,i (^0,2). The sequencers modulate the oscillating current to generate rf pulses. Typical pulse widths are of the order of 10 /us when addressing the two spins simultaneously and of the order of 1/Awo ~ 1 m s when addressing the spins individually. The rf pulses from the two sequencers are mixed and amplified. The subsequent process is similar to the heteronucleus case, except that two spins of homonucleus molecules are accessed by a single set of resonance circuit and receiver. 3.2.2. Hamiltonian in rotating frame We derive the Hamiltonian for homonucleus spins in the rotating frame, following the procedure similar to the heteronucleus case. If the condition AWQ S> J is satisfied, the system Hamiltonian in the rotating frame takes
144
the form H0 = hJ {Iz ® Iz)
(55)
when we set u>TOt,i = wo,j = o>rf,i as before. This coincides with the Hamiltonian (52) for heteronucleus spins. For the case of cytosine in D2O, we find Awo ~ 800 Hz while J ~ 7 Hz and hence the condition Aw0 3> J is satisfied. The Hamiltonian HTt^ becomes more complicated even after eliminating terms rapidly oscillating with frequencies 2wo,i or (a>o,i + ^0,2)- We obtain HTfti = hwiti [cos0i (Ix ® I) + sin^x (/„ (g> I)] +hujhi [cos(Aw01 + fa) (I
(56) where use has been made of the fact that the ratio wo,i/wo,2 of the Larmor frequencies is very close to 1 for homonucleus spins. If the pulse width is long enough or if w\ti
7777 Fig. 4. NMR setup for 2-qubit homonucleus molecules with two oscillators for individual qubits.
145
the same form as Eq. (53) for heteronucleus spins. At the same time, if the pulse width is short enough compared to 1/J, the J-coupling effects are negligible during the pulse operations. These observations imply that a pulse sequence designed for heteronucleus spins works also for homonucleus spins if hard pulses are replaced by soft pulses with proper pulse widths.
3.2.3. Common rotating frame In literature, a Hamiltonian jfo.conv = -fiAw 0 {I®IZ)
+ HJ (Iz ® Iz)
(57)
is often used as the system Hamiltonian for homonucleus molecules with two spins in a rotating frame.13 We call Eq. (57) the conventional Hamiltonian. The above Hamiltonian is derived from the Hamiltonian
#0,conv = - ^ 0 , 1 (Iz ® / ) - hu>0i2 (I <S> h)
+ HJ (Iz
(58)
in the laboratory frame through a transformation to a common rotating frame. The transformation operator to frames rotating with a common frequency wo,i is U = exp(—iwo,i Izt) ® exp(—iw0,i Izt),
(59)
which is obtained from U in Eq. (48) by taking a common frequency wrot,i = <^rot,2 = WQ,I. Then the Hamiltonian (46) transforms into
H0 = UH0tf
-
i U ^ at
= - A w 0 (/ ®IZ) + J [h ® Iz) +
/ 0 0 0 0^ 0 0^0
oI o o \o o o o/
(60)
The time development operators calculated with Eq. (58) and Eq. (60)
146
are related as, exp
-tf*
dt
/l 0 0 0^ 0 f(t) -g(ty 0 0g(t) /(*)* 0
Vo o /(*) =
s(t)
o
• exp
\)
-il>
O.conv
e^*/2(cos^-i -ie e
,iAui0t/2
J
Ao>o
(61)
dt
fit
fit
sm
fi Y' w
where fi = Awo\/l + j ^ h ej = J/|Awo|- Therefore, in the case of t ~ 1/ J and the weak coupling limit (ej < 1), the time development operator calculated with HQ can be approximated by that with Ho.conv This fact justifies that Eq. (58) can be employed as an approximate Hamiltonian Eq. (60) in the common rotating frame. Let us consider the gate
{< UE(6) = exp(-i9
IZ®IZ)
= \
i0/4
0 eie/4
0 0 0
0 0
0 0 eie/4
0
0 0 0
\
e -id/'
7
(62)
to show that the conventional Hamiltonian leads to inconvenient consequences. The two-qubit gate UE(TT) has been employed to implement a control-not gate along with one-qubit operations. 1 We consider a gate which is obtained by leaving the system with no applied rf pulses for a period t. If we take our Hamiltonian Ho of Eq. (55), the time development operator
Uj(t) = exp
"ijf*
dt
(63)
The distance between UE(K) and Uj(t) is easily evaluated as \\UEM
- Uj(t)\\ = 2V2\jl
- cos i ( J t - TT),
(64)
where the norm is defined as ||A|| = ^ti(A^A). The operator Uj(i) coincides with C/E at t = 2TT/2J. Note also that the distance 11t/"BC71") ~ ^ j ( t ) | | oscillates slowly in t with the period T = 8n/J.
147
If we take the conventional Hamiltonian Ho.conv of Eq. (57), in contrast, the distance between UE(K) and Uconvj(t) is ||£/EM
- UconvJ(t)\\
= 2V2Jl
- cos
(^Y^)
cos^(Jt
-
TT).
(65)
Therefore, if /fo.conv is employed, the operator Uconvj(t) fails to produce UE(TT) even at t — 2n/2J in general and the distance ||C/"E(7T) — Uconvj(t)\\ oscillates with the frequency Aw 0 /2. Although this inconvenience can be overcome by adjusting the phases of following pulses after UE{^) gates, we always employ the Hamiltonian (55) for its simplicity in this contribution. 3.3. Unitary
operations
for two
spins
Any quantum gate required for quantum computations can be decomposed into U(2) gates acting on individual qubits and CNOT gates between a certain pair of qubits. 16 The CNOT gate is a two-qubit operation in that a qubit, called a target qubit, is flipped only when another qubit, called a control qubit, is in the state |1) while the target qubit is left unchanged when the control qubit is in |0). Let us construct such unitary operations from the Hamiltonian H of Eq. (54) in the rotating frame. Our previous analysis confirms that it suffices to consider only the heteronucleus cases since the pulse sequences for homonucleus molecules are obtained from those for heteronucleus molecules by simply replacing hard pulses with soft pulses. One-qubit operations that we need are rotations by angle 6 around (cos 4>, sin
(66)
while the third term generates U^)
= I $1/4,(6),
(67)
where the indices 1 and 2 label the spins and U^,(6) is defined as in Eq. (14). Rotations around (0,0,1), namely UZti(6) and Uz^(6), can be constructed by employing a composite pulse or by redefining the axes in the rotating frame as explained in § 2.6.1. We assume here that the pulse widths are short enough so that the time-development due to the J-coupling term in H is negligible during the rf pulses. This assumption is safely satisfied for
148
heteronucleus molecules, for which these parameters are on the order of J ~ 100 Hz and wi,i ~ 25 kHz. In the case of homonucleus molecules, they are typically J ~ 10 Hz and wi^ ~ 100 Hz and thus one can still ignore the J-coupling term to a first approximation. We again denote the operations which rotate the i-th spin around the x-, y-, z-, —x-, —y- and — z-axes by an angle n/2 as Xi,Yi,Zi,Xi,Yi,Zi, see Eq. (32). Combining these elementary one-qubit operations with the two-qubit operation UE(0) we construct various gates that often appear in quantum algorithms. The first example is the Walsh-Hadamard gate H2 of two qubits (I H2 = HX®HX
=
1 1 1-111 1 - 1 V1-1-1
1 \ 1 1 1 /
(68)
It is constructed from H\ in Eq. (35) as H2 = {iYZZ) {iYZZ) =
(69)
—Y1Z1Z1Y2Z2Z2.
The second example is the CNOT gate, which is constructed by combining Uj (t) of Eq. (63) with one-qubit operations as
f^CNOT12 =
/1000\ 0100 = e l 7 r / 4 Z 1 Z 2 X 2 [/ E ( 7 r)y 2 , 0001
(70)
\ooio/ where the spin 1 is the control qubit while the spin 2 is the target qubit. Similarly,
^CNOT21 =
/1000\ 0001 0010 V0100/
e^ 4 Z 1 Z 2 X 1 C/ E ( 7 r)y 1 ,
(71)
where the spin 2 (1) is the control (target) qubit. Note that the overall phase givr/4^ r e q U j r e ( j to make the gate an element of SU(4), is not observable.
149
3.4. Field gradient for two
spins
Here we discuss effects of field gradient on two-spin system. Field gradient modifies the system Hamiltonian Eq. (46) to Ho = -Jiwo,i(l + Vz)Iz ® I - Jkj0<2(l + Vz)I ® J, + fiJ ] T Ji
where V is the normalized strength of the field gradient along the zdirection. We assume that the sample extension along the z-axis is — LQ < z < LQ. In the frame rotating with frequency u>o,i for the i-th spin, the relevant Hamiltonian is H0 = —huo.iVz IZ®I-
I ® Iz + hJIz ® Iz.
HLJO^Z
(73)
After application of the field gradient for a time interval r, the state is unitarily transformed by the operator UG(z) = exp[-i(-VuJotizIz
® I - Vu0,2zl ® Iz + JIZ ® IZ)T\.
(74)
The density matrix p in the rotating frame is a 4 x 4 Hermitian matrix. After application of the field gradient, the density matrix has ^-dependence
pG(z) = UG(z)pUG(z) / pu * *
* e-iAu,0TVzm
P22 e
iAworVz
p32
* \ ^
(75)
P33
V*
P44/
where * are terms containing ei2wo,iVzr o r ei(w0,i+w0,2)VzT xhese oscillating terms vanish after taking average over the sample coordinate z. A typical field gradient is V = 10 mT/m, for example, and its duration is r = 1 ms. The sample length is 2LQ = 4 cm in our case. In the case of heteronucleus molecules, u>o,i — Wo,2 and u>o,i are on the same order of magnitude, and thus the density matrix averaged over the sample is approximated by fpn 0 0 0 \ 0 p22 0 0 DGp = 0 0 p33 0
\ 0
0
0
p4J
(76)
150
while for homonucleus molecules, pG takes an approximate form
(Pn
0
0
0 \
0
P22 /°23
0
0
P32 P33
0
(77)
DG'P
\ 0 0 0 p4J The off-diagonal components remain in the latter case since | A W O | £ V T -C 1 for a typical choice of the parameters. Therefore, application of a pulsed field gradient works as a nonunitary transformation which eliminates most (in fact, all for heteronucleus molecules) of off-diagonal elements of the density matrix.
4. Density matrix for two spins 4.1. Thermal
state
4.1.1. Heteronucleus molecules The density matrix in a thermal equilibrium state is given by (see Eq. (5) for one-qubit molecules)
e-(H0/kBT)
Pth
tr(e-( H °/ f c s T ))
;•"
>I +
Ikuo,I kBT
/ z 5 D / + /w0,l
h 4
8kBT
$^
/ 0
j,
kBT +
0 0 0
WQ,2
0 0 0 0 w0,i - wo,2 0 0 0 — Wo,l + ^0,2 0 0 —0,1 —
\
w
0,2/
(78)
where H0 is given in Eq. (46). We can safely drop the term that contains J since WQ,J ^> J- Note that the matrix /9th is normalized as tr pth = 1-
151
4.1.2. Homonucleus Molecules In the case of homonucleus molecules, the density matrix is further simplified as 1
TO. T , ^ o . i , - _ , - . fiwo,i(l + Aw/w 0 ,i)
/100 0 \ 000 0 l/Sl-H*"0'1 4fcBT 0 0 0 0
r
_
T
(79)
Vooo - 1 / where use has been made of the condition Aw/wo,i <£ 1. 4.2. Pseudopure
state
The density matrix for n-spin molecules in NMR can be generally written
(80)
+ A/9
where "®n" denotes the n-th tensor power. The first term (J/2)®" represents an isotropic mixed ensemble, in which all spin states appear with equal probability. The second term Ap represents a deviation from the uniform ensemble. See also Eqs. (78) and (79). A unitary transformation U acts on the density matrix as <&n
+ Ap t/t
UpU* = U +
UApUl
(81)
This implies that only the deviation from the isotropic ensemble has relevance in time development of the system and the isotropic term {I/2)®n does not contribute to NMR signals. The mixed state Ppps
—
\ 2
+ adiag(l,0,...,0),
(82)
2«-l
is effectively equivalent to the pure state |00... 0). The real parameter a is of the order of /JWO,I//:BT. The non-vanishing component is not necessarily the first one. A quantum system whose state is exactly known is in a pure
152
state. In other words, a state is a pure state if and only if the rank of the density matrix is unity. Therefore, we treat the state Eq. (82) as if it were a pure state and thus we call it a pseudopure state. For a quantum computation the system should be initialized to be in a fiducial pure state. However, no unitary transformation p t—» UpU^ changes the rank of p, see Eq. (81). Hence it is impossible to get the pseudopure state from a thermal state by a unitary time development. Therefore non-unitary transformations are required to prepare pseudopure states.
4.3.
Initialization
A naive method to produce a pure state would be to cool the system under consideration. When the thermal energy fcsT becomes much smaller than the energy difference between the ground state and the first excited state, the system is definitely in the ground state. However, this method is not applicable to liquid-state NMR, since it usually works at room temperature and there the thermal energy is much larger than the Zeeman energy of a nucleus. Therefore we need a different method to produce a pseudopure state. Nonunitary transformations to produce a pseudopure state are classified into three categories: temporal averaging, spatial averaging, and logical labeling. Temporal and spatial averagings are based on the linearity of quantum mechanics. Suppose that there are N initial density matrices pi^. Let the same unitary operator U act on them. Then it yields N output density matrices po,i as Phi —* Po,i-
(83)
Quantum mechanical linearity guarantees that
£po,i =
tffe>/,iW.
If J2iPi,i 1S proportional to a pseudopure state ppps, portional to UpppSU^.
(84)
then J2iPo,i is pro-
4.3.1. Temporal averaging A suitable pulse sequence in NMR realizes a cyclic permutation of diagonal elements of the thermal density matrix Pth =
fl\®2 x
+diag(a 1 i,o 2 2,a 3 3,a44),
(85)
153
see Eq. (78). Two CNOT operations permute the diagonal elements as
Pth
(^CNOTl 2 £ / c N O T 2 1
*
+ d i a g ( a n , 044,022,033),
Pi
(86) t/cNOT2lfcNOT12
Pth
*
•diag(aii,a33,a44,a22).
P2
Note that the element a n is left invariant under these transformations. We obtain, after averaging over the density matrices, 1 , Ppps = ^{Pth + Pi + Pi)
I\v' 1 2J + o( a 22 + a 3 3 + 044)
/1000 0100 00 10
\oooi - ^ ( 3 a n - 022 - 033 - 044)
^1000^ 0000 0000
Voooo/ 1 + x ( 0 2 2 + 033 + 044)
+ o ( 3 a H ~ a 2 2 - 033 - 044)
/1000^ 0000 0000
(87)
Voooo/ This is a pseudopure state corresponding to a pure state |00). In the case of homonucleus molecules, (3an — 022 — 033 — 044)/3 = (2/3)(HuJoti/4kBT). In the case of homonucleus molecules (79) we can take temporal average more easily by taking into account the structure of the thermal density matrix Pth
fiwo, i (Iz
(88)
which implies that the thermal state itself may be regarded as a mixture of two states. Prom this observation, we find that ppps may be constructed with less
154
number of gates, as follows. Pth
Pth
f^CNOT12
UcNOT21
Note that 1 [(72 ® 7 + J ® J,) + (I 2 ® I + 2IZ ® 72) + (21* 0 J z + I ® 72)] 1 3
/3 0 0 0 \ /1000N 0 - 1 0 0 4 0000 0 0 - 1 0 ~~ 3 0 0 0 0 \0 0 0 - 1 / Voooo/
./®2
Therefore, PPPS = 3 (Pth + Pi + P2)
4 fi^o : 3 4kBT
/1000\ 0000 0000
(mod/® 2 ).
(89)
Voooo/ Here we omit irrelevant terms that are proportional to I®2. The above construction is called a product operator approach. 8 For two-qubit molecules, the number of initial states required to construct a pseudopure state by the product operator approach is three, which is the same as the necessary number required for cyclic permutation approach. However, the product operator approach provides twice larger signal than the cyclic permutation approach. Moreover, its preparation is easier (two CNOT gates in total) than the cyclic permutation approach which demands four CNOT gates. When the number of qubits is more than three, the product operator approach is more advantageous. For example, three initial states are enough for the product operator approach even for threequbit molecules, while the cyclic permutation approach requires 2 3 — 1 = 7 initial states. 4.3.2. Spatial averaging Spatial averaging approach employs pulsed field gradients to prepare piti inEq. (84).
155 We consider here homonucleus molecules first. A state Pth
>
-~^ I h 4fc B T
,/ + !/(
(90)
is prepared by a one-qubit operation and the field gradient (spatial averaging). We omit irrelevant terms that are proportional to (7/2)® 2 hereafter. Subsequent operations yield £>G2y3x/2,i(T/4)C/ E (7r)C/o, 1 (7r/4)
fiq>0[1
~'
4fcBT
, J
-/ ® I z (91)
'+27'
/1000N _ fi^o.i ' 0 0 0 0 4fcBT 0 0 0 0
(mod/82).
Voooo/ as promised. If heteronucleus molecules, such as ployed, the first operation is replaced as £>Gl^/2,2(7T/3) where rj satisfies Tkoi'0,1 4fc B T
WQ,I COST/ ^0,2
4fcBT
=
2WQ)2-
7®/s
13
C-labeled chloroform, are em-
(92)
£>Gl^/2,lM,
Then,
DGIU„/2,I(V)
hujp,2
(2IZ®I
+ I®IZ),
(93)
and the same second operation is applicable to generate a pseudopure state. There is another approach for equalizing populations 17 with Deq =
DGX1X2UE(TT/2)Y1Y2UE(n/2)X1X2.
(94)
Applying this we obtain ^0,1
4fc B T
T
^
^ 0 , 2 T/C> T Dm fi(w0,l + ^ 0 , 2 ) ,T I ®Iz > 77—^ \lz 4fcBT 4fcBT
(95)
after averaging over the sample molecules. The same operations are applicable once the populations are equalized. General procedures for arbitrary number of qubits are proposed by several groups. 18 ' 19 The method discussed here is a two-qubit version of Sakaguchi, Ozawa and Fukumi 18 which does not require an ancilla spin. In contrast, a method proposed by Sharf, Havel and Cory 19 requires an ancilla spin.
156
4.3.3. Logical labeling Logical labeling approach 20 to create a pseudopure state is understood by recalling the definition of a "cold" state. At absolute zero temperature, all the spins of a molecule align in the state |f). Suppose we have a molecule with N spins. If one can rearrange the populations with one- and two-qubit operations so that TV — M of iV-spins are aligned to | f) under a certain spin configuration of the other M-spins. In this case, N — M-spins can be considered at "0" K and it can be taken as a pseudopure state. Let us consider a more concrete case of molecules with 3 homonucleus spins. The thermal state of these spins is, pth oc diag(3,1,1, - 1 , 1 , - 1 , - 1 , - 3 ) .
(96)
By using a cyclic permutation of the populations, which is realized with two CNOT gates, pth is converted to, diag(3,1,1,1, - 1 , - 1 , - 1 , - 3 ) = Iz ® diag(3,1,1,1) /1000\
Voooo/ where we have dropped (7/2)®" as before. Therefore, when the first spin is |t), the other two spins are considered at "0" K and thus this subspace is considered to be a pseudopure state. Suppose we have an iV-qubit homonucleus molecule. The maximum number of qubits 7Vm in a pseudopure state by logical labeling is found as follows. Let 7Vth be the maximum number of equally populated states in thermal equilibrium. Then Nm is given by the integer which does not exceed log2(A^th + !)• In the 3-qubit example considered above, JVth = 3, namely there are three l's (or — l's) in the density matrix, see Eq. (96) and hence we find Nm = 2. When N is even, we find iVth = ( ^ 2 ) = N\/(N/2)\ since Nth is the number of states in which N/2 | | ) ' s and N/2 | | ) ' s are involved. On the other hand, when N is odd, 7Vth = ((Ar_m/2) because i\Tth is the number of states in which (N + l)/2 of ||) and (N - l)/2 of ||) are involved. We do not discuss the logical labeling approach in the rest of this article.
157
4.4. Quantum
state
tomography
The density matrix p of two-qubit molecules in a rotating frame is parametrized as
/ ^=2®2
+
an
an + ibn au + ibu au +
012 - ib\2
a22
013 - ^ 1 3 «23 - i&23 \ a j 4 — ibu
ibu\
a23 + ^ 2 3 024 + ^ 2 4 ^33
(98)
&34 + ^ 3 4
CL2A — i&24 a 3 4 — *&34
<^44
/
Note that 1/2
Y^
CijIi^Ij,
(99)
where IQ = 1/2. Since coo = 1 for normalization, the number of free parameters is 4 x 4 — 1 = 15. By measuring all aij and 6^, we know the state completely: this is called quantum state tomography, see also § 2.4. A state of a multi-qubit system develops in time driven by J-couplings among spins. Therefore, the actual final state after operation of a quantum algorithm (63) in the rotating frame is
p(t) = Uj(t)'PU\(t).
(100)
The density matrix p(t) in the laboratory frame is
p{t) = tfp(t)U = exp [iw0,i t(Iz ® I) + iuj0t2 t(I (g> Iz)] p(t) • exp [-iw 0 ,i t(Iz
(101)
158
Then the ^-component of the magnetization in the laboratory frame is tr \{<7X ®I + I® crx)p(t)} ai2 cos ( w0,2 - -z ) t + a 34 cos ( w0>2 + — J t +a 1 3 cos I w0,i - — J i + a24COS I w0,i + — ) t\ 612 sin ( ujQfl - - J t + 634 sin ( w0,2 - 77 J t (102)
+613 sin ( w0,i - — J t + 624 sin ( w0,i - ~z I *
Therefore ai2,034,013, a24 are measured from the real part of the spectrum while 612,634, 613,624 from the imaginary part. See, § 2.5. Suppose next that a 7r/2-pulse Yi around the y-axis in the rotating frame is applied to the spin 1. The density matrix after this operation is YlPYl
(103)
The x-projections in the laboratory frame after this 7r/2-pulse is »12 VV— 2 2
014 _ Q23
^)cOsL,2-^U
2 ~ 2
/<2i '12 . 014 , «23 . 0 3 4 \
/
, J\ ,
+ V -2-2 + ^ - + -2~ + ^ - J C O S ( a ; o - 2 + 2 . fan +
+
a33\ / COS
623
612
1
/a22
1
a44\
+
( COS
, J\ '
0 1+
p- " 2 J * (TJ - T) r - 2 J *.
{-Y--Y)614 2
J\
U
- +1 -+
634
T
|SmlWo,2-2|i
+624 sin ( w0,2 + - j i - &13 sin ( w0,i - - J t 612
614 623
•T"T
+
T-T
634 \ . ( sin
. J , , W 0
-
1 +
2
U
(104) We extract from (104) the information 011—033,
022—044,
Oi4+<223,
614 — 623-
Similary, measurement after a Y2-pulse provides the information On-a22,
033-044,
014 + 023, 6 14 + 623
159
while measurement after a Xi-pulse provides the inforamtion ^ 1 4 + ^23J
a
l l — a 33i
^22 — 044,
^14 ~ 023-
Therefore, all a^- and b^ can be determined experimentally. Note that there are other combinations of measurements to yield {a^} and {bij}. Let us consider a simple case in which p = diag(an,a22,a33,a44). Application of a 7r/2-pulse Y\ leads to the following ^-projection: tr
(ax ® /) YipYf] = ± an cos ( WQ,I - -z ) t + a22 cos I w0,i + — ) t - a 3 3 cos f w0li - — j t - a 44 cos f w0,i + — j t (105)
Since not few quantum computations result in a density matrix in which only one diagonal element is 1 while all the other elements being 0, the above method provides a quick way to confirm the result of computation. Equation (105) tells us that the state of the spin 1 can be read from the signature in front of a cos function; the signature + (—) corresponds to IT) = |0) (11) = ID)- The state of the spin 2 is read from the peak position in the spectrum; a larger (smaller) frequency shift corresponds to ||) (|D) for the spin 2, assuming J > 0. 5. Quantum computation We have demonstrated so far that we can perform one- and two-qubit operations and measure the quantum states of the system using Hamiltonian formulation. We are now ready to perform quantum computations with twoqubit molecules. We examine here two most popular quantum algorithms; the Deutsch-Jozsa (DJ) algorithm and Grover's database search algorithm. First, we will briefly review these algorithms and show the results of our experiments which reproduce the results reported in literature. Then, we will introduce our own contributions. 5.1. Deutsch-Jozsa
algorithm
5.1.1. Background The DJ algorithm is a quantum algorithm proposed by Deutsch and Jozsa in 1992.21 It is one of the first algorithms which take advantage of quantum nature, superposition and interference, in computation.
160
output f input
O •
f 2
1
O O
f
3
•
o
•
•
f 4
•
o
Fig. 5. Deutsch-Jozsa problem for one input variable. The task is to determine whether unknown fk is constant (/i or {2) or balanced (/3 or f$).
Suppose there is an input of one bit, 0 or 1, as shown in Fig. 5. The white ball represents 0 and the black ball represents 1 for example. There are four boxes corresponding to four distinct functions, /i(0) = 0,
/i(l)=0,
/ 2 (0) = 1,
/ 2 (1) = 1,
/s(0) = 0,
/3(1) = 1, / 4 ( i ) = 0.
/4(o) = 1,
For example, when a black ball (= 1) is thrown into the box 1 (function /1), a white ball (= 0) comes out. The task is to determine whether unknown fk is constant (/1 or / 2 ) or balanced (/ 3 or f^, for which half of the output is 0 and the rest is 1) by throwing balls into the unknown box as few times as possible. This task is achieved, quantum mechanically, by throwing a single ball which is a superposition of white and black balls. In a general DJ problem, we consider a function which has n bits input and one bit output / : {0,1}" ^ { 0 , 1 } ; If f(xi,x2,...,xn)
(xi,x2,...,xn)^x.
= 0 or f(x1,x2,...,xn)
= 1 for all {xi,x2,...,xn)
(106) e
161
{ 0 , l } n , the function / is called constant. If f(x\,X2,--.,xn) = 0 for half of {0,1}™ and f(x\,X2, •. • ,xn) = 1 for the rest, the function / is called "balanced". Note that not all functions / : {0,1}" —* {0,1} are classified into these two classes. Our task is to determine whether / is constant or balanced assuming that / belongs to one of these two classes. It takes at least 2 n _ 1 + 1 steps to tell if a given / is balanced or constant when a classical algorithm is employed. If, instead, the DJ quantum algorithm is employed, we need to evaluate the function once for all. The DJ algorithm is shown in a quantum circuit form in Fig. 6. The XOR (exclusive-OR) operation can be realized with {/CNOTI2 and E/CNOT21 as shown in Fig. 7.
X
|0> - / * -
y
n\ [1/
Hfn
rr®n
Uf
f:\x,y)-*\x,y®f(x))
H-
-"1
T
T |03>
Fig. 6. A quantum circuit implementing the DJ algorithm. The symbol / n represents a set of n qubits. H\ is the Hadamard gate. The symbol ffi is the XOR (exclusive-OR) operation.
C^CNOT12
spinl
\x)
spin 2
\y)
|x>
-©-
\x®y)
-$"
\x®y)
t/cNOT21
Fig. 7.
spin 1
|a;)
spin 2
\y)
\y)
The XOR (exclusive-OR) operation can be realized with [ / C N O T I 2 and
t^CNOT21-
162
The quantum state develops at each step in Fig. 6 as 1 1 « = |0} 8 "|1>
l 1,
* "Jj,.^ ( ~^~ ) ,6(0.1)- ^
^
V
z£{0,l}" x £ { 0 , l } "
Z-<
2n
Z^
^
zG{0,l}nxe{0,l}"
A/2 V
The identity H®n \x) = ^ Z ( - 1 ) I Z \z) /y/2" has been used in evaluating |
(107)
E H^xe{o,i}"
If / is constant, namely, if / = 0 or / = 1, Eq. (107) becomes ± 1 . On the other hand, if / is balanced, the coefficient vanishes since (.!)/(») / -> xe{o,i}"
2
n
(_i)0 J
xef-HO)
= 1-1=0
2
n
(_i)i ^—-^ xef-1^)
2"
(108)
due to interference among states. Therefore, if the probability to observe the state | 0 , 0 , . . . ,0) is 1, then / must be constant. The function / must be balanced if the state | 0 , 0 , . . . , 0) is not observed.
5.1.2.
Implementation
The DJ algorithm (Fig. 5) is implemented with a liquid-state NMR. The NMR pulse sequences are shown in Table l. 8 The Hadamard gate is constructed with a 7r/2-pulse. The gate Uf is constructed with several pulses and the J-coupling, i.e. U-E{6). Results of the DJ algorithm is summarized in Table 2.
163 Table 1. The pulse sequences for the DJ algorithm for one-bit functions that produce the unitary matrices in the right column. spinl |0) spin2 |0)
Y Uf
Y
/i(i)sO
l/E(7r/2)
UE(TT/2)
J - X2
h{x) = 1 £/ E (7r/2)
- X2
0 i 0 0> t 0 0 0 0 0 0 i OOiO,
(7 E (7r/2)
-
h{*) =x I/BW fi{x)
= 1 -x
1
r^HZF -
UEM
HxHZHZ}
(-1)
1/4
1 0 0 0
0 0 0> 100 0 0 1 0 10/
'0100^ 10 0 0 0 0 10 , 0 0 0 1/
Y — X
5.1.3. Experiments Data were taken at room temperature with a JEOL ECA-500 spectrometer, whose hydrogen Larmor frequency is approximately 500 MHz. 22 We used 0.6 mL, 23 mM sample of cytosine in D2O. 23 Spins of two hydrogen nuclei in a cytosine molecule work as qubits. The measured coupling strength is J/2n = 7.1 Hz and the frequency difference is Aui/2ir = 765.0 Hz. The transverse relaxation time T2 is ~ 1 s for the both hydrogen nuclei and the longitudinal relaxation time T\ is ~ 7 s. It is known that the initial spin state to execute the DJ algorithm may be a thermal state. 24 The density matrix becomes diagonal after the DJ algorithm is executed. Therefore we need to apply one pulse for quantum state tomography as shown in Eq. (105). Expected signals are summarized in Table 3. Our results of quantum computations with cytosine in D2O are summarized in Fig. 8. The figures show Fourier transformed spectra of signals from the spin 1. Since the initial state is in thermal equilibrium, molecular spins
164 Table 2. Inputs and outputs of the Deutsch-Jozsa algorithm. The algorithm is carried out with the pulse sequences given in Table 1. |x) is the state of the spin 1 and \y) is the state of the spin 2. In can be seen that the outputs of the spin 2 gives f(x) © y. The phase of states is not considered. function fi(x) = 0
f2(x)
= 1
/3(x) = x
U(x) = NOT(x)
input \xy) |00) |01> |10) 111) |00) |01) |10) 111) |00) |01) |10) 111) |00) |01) |10) 111)
fj{x) 0 0 0 0 1 1 1 1 0 0 1 1 1 1 0 0
y ® fj(x) 0 1 0 1 1 0 1 0 0 1 1 0 1 0 0 1
output \xy) |00) |01) |10) 111) |01) |00) 111) |10) |00) |01) 111) |10) |01) |00) |10) 111)
Table 3. The expected signals from the resulting state of the pulse sequence in Table 1 followed by the reading pulse Y\ on the spin 1. The effect of relaxation is ignored. function /i h
expected signal cos(u>o,i — J/2)t + cos(u;o,i + J / 2 ) t cos(u>o,i - J/2)t + cos(u;o,i +
/3
- COS(CJO,I - J/2)t
+ cos(o>cu +
fi
— cos(u;o,i — J/2)t + cos(wo,i +
J/2)t J/2)t
J/2)t
are mixture of four states |00), |01), ]10), and |11). When the initial state of the spin 2 is |1), the algorithm fails to distinguish a constant function from a balanced function. In contrast, it works regardless of the state of the spin l. 24 Peaks with a larger frequency shift (the left peaks on each curve) in Fig. 8 correspond to an initial state in which the spin 2 is in the state |0) and successfully tell if fi is constant or balanced. On the other hand, the peaks with a smaller frequency shift (the right peaks) correspond to the initial state in which the spin 2 is |1) and hence fails.
165
5.8 ppm
? Fig. 8. Spectra of signals from the spin 1 after application of pulse sequences of the Deutsch-Jozsa algorithm and the reading pulse. The peaks with a larger frequency shift (the left peaks) are signals that result from the initial spin 2 state = |0). They tell if the function / ; is constant or balanced from the signature of the peak. T h e peaks are positive for / j and /2 (constant) and are negative for f$ and fi (balanced).
5.2. Field inhomogeneity
compensation
5.2.1. Application of ir-pulse pairs during J-coupling operation Molecules in NMR quantum computing are often under the influence of field inhomogeneity. This may cause an error in realization of UEIt is well known that an undesired effect caused by field inhomogeneity can be compensated by a series of hard 7r-pulse pairs, where the width of each pulse is on the order of 10 /xs. The best known example may be the CPMG (Carr-Purcell-Meiboom-Gill) pulse sequence.9 By applying this technique to UE, we replace Uj(t) with Uj*(t) = Uj(t/2) - 7T - Uj{t/2) - 7T,
(109)
where TT denotes a hard 7r-pulse around the rc-axis of one of the two spins, for example the spin 1, and time flows from the left to the right in the right hand side. According to Eq. (56), the above operation is described as UMt) = ul2}Uj(t/2)U?}Uj(t/2),
(110)
166
where U^\ = exp (-inlx
x exp [-i?r(cos(Aa;i/2)(/
® Iy))], (111)
U®\ = exp (-ITTJX <8> /)
x exp [—i7r(cos Au)t(I (g> 7X) + sin Autt(I ® 7^))]. We assume that the 7r-pulse is around the x-axis of the spin 1 in its rotating frame and is instantaneous. Thus the rotation axis for the spin 2 turns with the angular velocity Aw during Uj(t/2). The distance between Uj*(t) and UE{TT) is evaluated as before
H ^ E W - Uj.(t)\\ = 2V2J1 - cos ( ~ )
cos i (Jt - TT).
(112)
The distance does not vanish for any t since two conditions cos (Auit/2) = ±1 and cos | (Jt — TT) = ± 1 are not simultaneously satisfied in general. Moreover, the distance oscillates rapidly in t through cos(Awi/2). 5.2.2. Pseudopure state by spatial labeling We applied a 7r-pulse pair during the two-qubit entangling operation C/E(TT) in the pulse sequence which creates the pseudopure state |00) by spatial labeling, which was discussed in § 4.3.2. The entangling operation is realized by turning off rf pulses for a specified time. The experimental results are summarized in Fig. 9. The spectra in the left panel were obtained without a 7r-pulse pair. We should observe, in principle, only a single peak with a larger frequency shift, namely a signal from molecules in the state |00). A small peak at frequency with smaller frequency shift indicates the spurious state |01). Those in the right panel were obtained with a 7r-pulse pair applied during the entangling operation. We see that they are very sensitive to the entangling operation time t. When t was set at the correct value 70.3 ms, the spectrum exhibited a sharp peak at a frequency with larger shift and a small peak at a frequency with smaller shift. The sharpest peak obtained with the right value of t was sharper than those in the left panel. This result implies that a 7r-pulse pair improves quality of the pseudopure state. However, even a small deviation of t from the correct value distorts the spectrum considerably. The spectrum with t = 69.7 ms, for example, indicates that the created state is actually |01), not the desired |00).
167
6.0 ppm
5.8 ppm
Fig. 9. The spectra of the spin 1 in the pseudopure state. Spectra in the left panel were obtained without 7r-pulse pairs while the spectra in the right panel were obtained with 7r-pulse pairs applied during the two-qubit entanglement operation. The entangling time was varied from 68.8 ms to 70.0 ms in the left panel while from 69.3 ms to 70.6 ms in t h e right panel. The peak with a larger frequency shift (the left peak on each curve) corresponds to the state |00) while the peak with a smaller frequency shift (the right peak) corresponds to the state |01). The latter peak indicates an error in the pseudopure state preparation. The spectra in the right panel are sensitive to variation of the entangling time.
5.3. Optimal
Implementation
of Two-Qubit
algorithms
We need to find a control function 7(2) in Eq. (1) to implement a quantum algorithm f/aig so that Uh(t)} = t/alg-
(113)
It is shown25 that Ua\g can be decomposed into three matrices as C/aig = k2hh,
(114)
where k\,k2 G SU(2)
h(tx,ty,tz)
= Y\_ UuiU).
(115)
Here Uxx{t) = e-^1*®1* iJtI
Uvy(t) = e- y^y
= Y1Y2UJ{t)Y2Yl =
X1X2UJ(t)X2X1
(116)
Uzz(t) = Uj(t), are two-qubit unitary transformations which cannot be decomposed into a tensor product of one-qubit unitary transformations in general. The decomposition (114) is called a Cartan decomposition. Since the execution
168
time of ki is much shorter than that of Uj(ti), the total execution time is approximated as T = J2i=x,y,z tilt is possible to explicitly write down the Cartan decomposition of any U € SU(4). 25 ' 26 Let us introduce the "magic" basis 27 |*o) = ^ ( | 0 0 > + | l l » , |*1> = -^(|01> + |10», f |* 2 ) = _ ( | 0 1 ) - | 1 0 » ,
(117)
|* 3 > = ^ ( | 0 0 ) - | 1 1 ) ) . Under the basis change from the binary basis |00), |01), |10), |11) to the magic basis, a matrix U transforms as U —> UB = Q^UQ, where
« - *
/10 Oi Oi \l0
0 i \ 1 0 -10 0 -i)
(118)
The matrix Q defines an isomorphism between K = SU(2) <S> SU(2) and SO(4) by Q^kiQ £ SO(4) and is used to classify two-qubit gates. 27 ' 29 Some examples are shown in Table 4. Moreover, Q diagonalizes elements of the Cartan subgroup, viz Qt hQ = diag(e' e °, eldl, e102, e%9a). Useful examples are found in Table 5. T a b l e 4. T y p i c a l e x a m p l e s of Q t f c Q . Ui{6,4>, £) is a m a t r i x r e p r e s e n t a t i o n of a r o t a t i o n b y a n a n g l e £ a r o u n d t h e a x i s (sin 0 cos tp, sin Q sin cf>, cos 9) of t h e s p i n i. k
QtfcQ
( t/2(e,0,O
cos §
cos
sin
cos 9 sin §
\
— cos <j> sin 9 sin | cos § — cos 8 sin | sin 4> sin 8 sin | j — sin <j> sin 6 sin | cos 0 sin | cos § — cos
169 Table 5.
Typical examples of Q^hQ.
h
QtftQ /10001
Uxx{2-n/2J) = e-**'*®'*
t/w(27r/2J)=e--Wv
e--/4 I g J ° °
/1000' , - w W ° *0 0 \00Oi, 1000' OOiO 0 0 0 1,
Let U be a matrix to be decomposed. The matrix U in the magic base takes the form UB = Q^UQ.
(119)
If UB € SO(4), namely if U^UB = h, then U belongs to K and there is no need for Cartan decomposition. Therefore we assume UB £ SO(4). Then observe that UB = Q^UQ = Q^k2Q • Q]hQ • Q]kxQ = 02hD01,
(120)
where O* = Q^ktQ € SO(4) and hD = Q^hQ is a diagonal matrix. From UlUB = Ojh2DOu
(121)
we notice that U^UB is diagonalized by the orthogonal matrix 0\ and its eigenvalues constitute the diagonal elements of h2D. It implies that 0\ and h2D can be determined from the eigenvalues and eigenvectors of UQUBNote, however, that there is some ambiguities in the definition of ho when taking square root of h?D. It can be shown 28 that there always exists a solution in which tx > ty > \tz\, namely, the point (tx,ty,tz) is assumed to be in the Weyl chamber of su(4). 29 Then fci = QO\Q^ and h = QhoQ^ are found straightforwardly. Finally, 02 is fixed as 02 = UB^DOI)"1 and hi = Q02Q^ is obtained. 5.3.1. Examples As a first example, we consider a trivial case of the Walsh-Hadamard gate H2 for two qubits given in (Eq. 68). We need to know first if H2 is implemented with or without J-coupling operations. The solution is easily found
170
by calculating / 1 0 0 Q\ 00 0 1 QpH2Q = 0 0 - 1 0
(122)
Voi o o/ and UlUB = {Q^H2Q)T
(Q^H2Q) = J 4 .
(123)
It implies that H2 S SO (4) and is implemented without any J-coupling operations as we have already shown in Eq. (69). As a second example, let us determine NMR pulse sequence for the controlled-phase gate /100 0 \ 010 0 00 1 0
tfcp
(124)
Vooo-i/ The controlled-phase gate is employed to implement a CNOT gate along with one-qubit operations. 1 We find
UlUB = {Q^UcpQf
/-100 0 \ 0 10 0 ¥= h(<2fE/cpQ) = 0 010
V0
(125)
00-lJ
Therefore, the NMR pulse sequence to implement Ucp involves twoqubit operations, such as Uxx, Uyy, or Uzz. The eigenvalues of U^UB are 1, —1, —1,1 and the corresponding eigenvectors are ( 0 , 1 , 0 , 0 ) r , (1,0,0,0) T , (0,0,0,1) T , (0,0,1,0) T . Therefore, / l 0 0 0\ 0-100 h% = 0 0-10
(126)
\o o o \) and
/oioo^ Oi =
1000 0001
Vooio/
(127)
171
are obtained. Note that this is one of the possible choices. We choose it since ho takes a form
hD
/l 000\ 0 i 00 = 00 i 0
(128)
Voooi/ so that we can construct hp as hD =
e™'lUzz(-K/J)
(129)
using Table 5. Then, 02 should be
UB(hD0{)-x
02 =
/0 0 10\ 1 0 00 0 0 01
(130)
V0-100/ which is certainly an element of SO (4). The element k\ is easily obtained
fci = QOiQ]
/0 0 - i 0 \ 00 Oi = iO 00
Y1Y1Z2Z2.
(131)
Vo -i 0 0/ Similarly we obtain
/0 0
-i0\
0 0 0 i i 0 0 0
=
Z1Y1Y1Z2,
(132)
Vo —i 0 0/ where we used the notations defined in Eqs. (66) and (67). Combining the above calculations, we obtain the NMR pulse sequence to implement UCP as k2 hkj. = {ZxYxYxZ*) (f^*Utz(n/J)) =
(Y1Y1Z2Z2)
(133)
-jir'iZ1Z2Y1Y1UE{n)Y1Yi,
where we employed the rules given in Eq. (33) to simply the pulse sequence.
172
5.4. Grover's
database
search
algorithm
5.4.1. Background Suppose there is an unstructured database of N entries. One of the entries satisfies a given condition but we do not know which one is the desired entry. The database search is a task to find the entry that satisfies the condition. Classically, the only possible method to find the entry from the unsorted database is (1) Pick up any entry in the database. (2) Check whether it satisfies the condition. (3) If it satisfies the condition, it is the entry we are looking for. If not, repeat (1) and (2) till the desired entry is found. The condition is defined by a function f(i), which is called an oracle, such that f(i) = 0 or 1 for each entry i and f(w) = 1 is realized only for one desired entry w. The target entry w is called a "file" to be found. In the worst case N — 1 times queries (the steps (1) through (2)) are necessary and the average number of necessary queries is N/2. In 1996 Grover 30 proposed a quantum algorithm for the database search, with which one can find the desired entry with as small as ~ VN queries. In quantum information processing the set of TV data entries is represented by an ^-dimensional space spanned by orthogonal basis representing each entry. Then, a database search problem is reduced to a task to find the state \w) that represents the entry satisfying the condition f(w) = 1. Grover's database search algorithm 30 consists of the following procedures: (1) Prepare the initial state \s) which is a superposition of all the entries |t>:
i
If JV = 2 m , then \s) can be easily prepared by the Walsh-Hadamard gate as \s) = H®m |0)
(135)
173
(2) Let a unitary operator G=(2|S>(S|-j)m-l)/W|i>(i|j =
(136)
(2\s)(s\-l)(l-2\w){w\)
act n (~ y/N) times on \s). Here / is the identity operator. The action of G on \
= (2|a>(a|-/)(|0)-2cU(|t(;)) = 2 \s) (s \
i,j
i,j,k
Y/(2c-ci-^)\i)+2cw\w)
=
i
~ 5Z(2c -Ci)\i)
+ (2c + cw)\w).
(137)
Here, c = ( ^ cfi/N or the mean value of Cj. (3) After n = [7r/40] iterations, where [ ] stands for the integer part and sin 0 = y/l/N, cos 8 = y/l — 1/N, the probability amplitude localizes at \w), while all the other amplitude being negligible. Therefore the measurement of the state gives \w) with a probability very close to 1 and the database search is completed. Note that 6 ~ 1/vN for a large enough TV and the algorithm requires only n ~ y/N steps. Grover's algorithm can be understood more visually as follows.31 Define a vector \n) =
£N>,
(138)
which is orthogonal to \w). If we write the initial state \s) with \n) and \w), \s) =cos6\n)+sm6\w).
(139)
174
Then, action of G gives G{a\n)+P\w))
=G =
/cos26> -sin2<9\ (a\ \**29 cos20 ){p)-
(140)
Therefore, Grover's iteration is regarded as repetition of rotations by an angle 29 from \s) towards \w) in the 2-dimensional subspace spanned by \n) and \w). The number of necessary iterations is n = n/AO ~ iry/N/4. The algorithm for N = 2 2 = 4 entries is summarized in Table 6. The case of N — 4 is special in the sense that only one Grover's iteration gives the destination file. Table 6. Quantum circuit for Grover's algorithm for N = 2 2 = 4 entries. Rij is a selective inversion and Uij is a total unitary operation representing the algorithm. H2R00H2 2|*><*|-/
10 10
Rij
0
o-io 0
0
o
0 - 1 0
o o o - i
-10 0 0 0 10 0 0 0-10
ooo-i, 0 0 10^ -10 0 0 0 0 0-1 0 - 1 0 0 ,
5.4.2.
1 ) ) )
11
0 0 0 -1 0 0 0 1 0 0 0 - 1
) 1 0 ) 0 0 100 1 0 - 1
\ j I /
0 \ -1 I OJ 0 /
/-l
0 0 0 - 1 0 10 0-1 \ 0 0 0 / 0 [ 0 0 \-l
0 - 1 00
0\ 0 0 1 /
0 1\ 0 0 1 1 0 I 0 0 /
Implementation
Chuang et al.32 employed the pulse sequences shown in Table 7. Note that the matrices produced by their pulse sequences do not agree with those of actual Grover's algorithms: The relative phase of their matrix elements do not reproduce those in Table 6, although the difference does not affect NMR measurements. We present our pulse sequences in Table 8, which are further simplified from the pulses of Chuang et al by employing the rules in Eq. (33). For
175 Table 7. Pulse sequences employed in Chuang et al. to implement Grover's algorithms. The pair (A, B) should be substituted by (X,X),(X,X),(X,X), and (X,X) for w = 00, 01, 10, and 11, respectively.
Ui,
Table 8. Simplified pulse sequences for Grover's algorithms. The pair {A, B) should be substituted by (X,Y),{X,Y),(X,Y), and (X,Y) when w = 00, 01, 10, and 11, respectively.
HEHU00
01
10
11
example, we simplify Uoi as U0i =
X1Y1X2YUE(n)XlY1X2Y2UE(iT)Y1Y2
=
X1Y1X2YU^{-K)Z1X1Z2X2U^{-K)Y1Y2
=
Z2Z1Y1X1Y2X2UE(Tr)X1X2UE(n)Y1Y2
=
Z2Z2Z1Z1Y1Y2UE(Tr)X1X2UE(Tr)Y1Y2.
(141)
The factors Zi in the end of the sequence are irrelevant and can be removed to yield
Voi =
Y&UvWX&iUvWYM.
(142)
Although we have reduced the number of pulses by the above simplification, the role played by each step in Grover's iteration becomes less clear then. 5.4.3. Algorithm acceleration Here we briefly describe our contributions toward reduction of execution time in quantum computation. See the references 33 ' 26 for further details.
176
As discussed in the previous section, we successfully reduced the number of pulses required to implement Grover's database search algorithm. Although the reduction in the number of pulses is quite helpful to reduce the gate operation errors, the total execution time is hardly changed since the number of the most time-consuming gates, UE(IT), is unchanged. Recall that J-coupling is the slowest process in NMR quantum computing. Here we show how to attain the optimal execution time and reduce the number of gates. Note that the shorter execution time is important to overcome decoherence. The strategy, which we call "quantum algorithm acceleration" was originally proposed for fictitious Josephson charge qubits. 34 ' 35 For a realization of a quantum algorithm U, we have to implement an n-qubit matrix which represents the quantum algorithm. Note that there is no necessity to implement individual elementary gates often employed to construct U. This matrix may be directly implemented by properly choosing the control parameters in the Hamiltonian. The variational principle tells us that the gate execution time can be reduced, in general, compared to the conventional construction with elementary gates since the conventional gate sequence is one of the possible solutions. The quantum circuits for two-qubit Grover's algorithm is shown in the first row of Table 9. We obtained the optimal Cartan decomposition for Grover's algorithm by employing the method outlined in § 5.3. The result shows that the NMR execution times T = Yli=i U f ° r m a discrete series as T = ( n + 1)/J,
n = 0,l,2...
(143)
for all the "target files" |00), [01), 110), |11). The minimum execution time is 1/J, which shows that the conventional pulse sequence in the first row of Table 7 is already time-optimal. The initial state for Grover's algorithm is a pseudopure state prepared by applying cyclic permutations (CP) on a thermal state population as discussed in § 4.3.1. We merged the cyclic permutation gates UCP and £/CP 2 with Grover's search algorithm as shown in the second and third rows of the upper half of Table 9. We obtained remarkable results when we optimized the pulse sequences for the combined unitary gates. The optimized quantum circuits are shown in the lower half of Table 9. Although we added extra gates to Grover's algorithm, the execution times became shorter than original Grover's search algorithms. The experimental setup is the same as in the case of the Deutsch-Jozsa
Table 9. Pulse sequences of Grover's algorithm picking u p t h e "file" |01). T h e initial s t a t e |00 cyclic p e r m u t a t i o n s of t h e s t a t e populations. Pulse sequence Conventional
-EH U E W -EH UQIUCP
-mi
UEW
ym<EH
£^01
^CP2
UEM
-HH3-.
UEM
—m-\—HEH^I— HIHEH
L&4IH
HfH^r-^ Ue(ir
Optimized Uoi
4Eh
C/EW
-mUoiUcp
UoiUCP2 f
iVote: T h e number of pulses in t h e optimized pulse sequences is reduced from 18 in our p a p e r 3 phases of t h e elements in t h e matrices realized with these pulse sequences do not coincide with t t h e end of each sequence have been dropped as in Eq. (142).
178
algorithm. The sample molecule is cytosine solved in D2O. The initial state is the thermal state. We performed the NMR pulse sequences listed in Table 9. The expected signals from the sample are given in Table 10. Table 10. Expected signals which will be observed after performance of Grover's algorithm followed by the reading pulse on the spin 1 (Yi). The sum of signals (temporal averaging) is taken to define the pseudopure state. operation U01
signal \ cos(w 0 ,i - J/2)t + \ cos(u>o,i +
UOIUCP
cos(o>o,i +
UoiUCP2
—5 cos(wo,i — J/2)t + I cos(o;o,i +
Sum of the signals
2cos(u;o,i +
J/2)t
J/2)t
J/2)t
J/2)t
We applied pulse sequences listed in Table 9 on cytosine in the thermal state and observed the spectra shown in Fig. 10. Signals from spin 1 were measured and Fourier transformed. The left panel in Fig. 10 shows the results obtained with the conventional pulse sequences (the upper half of Table 9), while the right panel shows the results obtained with our accelerated algorithm (the lower half of Table 9). In both cases, we obtained signals that are in agreement with the theoretical expectation. This result proves that the quantum algorithm acceleration works correctly. The sum of spectra of three performances, {C/01, UQIUCP, ^ O I ^ C P 2 } m Fig. 10, corresponds to that of Uoi [00), where |00) is the pseudopure state created by temporal averaging. 5.4.4. Warp-drive quantum computation We extended the idea of quantum algorithm acceleration further and actively reduced the execution time by adding an extra gate called the "warpdrive" gate. 26 When a permutation matrix W of the binary basis vectors is multiplied on the unitary gate C/aig, W may send Ua\g to a point WUa\g near the identity matrix / so that it takes a shorter time to follow the time-optimal path connecting / with WUa,ig than that connecting I with t/aig as illustrated in Fig. 11. We call this technique "warp-drive" since the reduction in the execution time takes place instantaneously as the extra gate is introduced. We have "warp-driven" Ua\g to WUaig by adding W to £4ig-
179
Un UQIUCP
W* sum 6.0 ppm
5.8 ppm
Fig. 10. Fourier transformed signals of the spin 1 after performance of Grover's database search algorithm. The left panel shows the results of the conventional pulse sequences, while the right panel are results of our accelerated pulse sequences. In both cases, the observed signals are in good agreement with the theoretically expected spectra listed in Table 10.
Fig. 11. A conceptual picture showing the action of the warp-drive gate W. The sphere represents the compact SU(4) group manifold. The warp-drive W sends a unitary matrix f/ a l g to WU^ig. The curves connecting these matrices represent time-optimal paths. The product unitary matrix WUa\s is reachable from the identity matrix 7 in a shorter execution time than U,alg'
The extra gate W which shortens the execution time must be simple enough so that we can deduce the output of f/aig|0) efficiently from the output of VT£/aig|0) using classical computation only. Here |0) denotes the n-qubit fiducial state |00.. .0). We call such gates warp-drive gates. Since
180
a matrix Ua\e is an element of a compact group U(2 n ), there always exist such warp-drive gates which will reduce the execution time. Let us show a more concrete example in the case of Grover's algorithms. When Uio is Cartan-decomposed, the optimized pulse sequence is obtained as shown in the first row in Fig. 5.4.4. Note that the number of UE{^) gate is two and thus the execution time is 2n/J. Then, we add the warp-drive gate (Ucp = t/cNOTi2 ^CNOT2i) to Uio and obtained the pulse sequence shown in the second row in Fig. 5.4.4. Note that the number of C/E(TT) gate is one and thus the execution time is halved compared with that of UioU10 :
v
V
X —
A UE(IT)
X
A
X —
UcpUw
1
C/E(TT)
Y
i
U-n/i(n)
1^/4(71-)
X UE(ir)
h ^W Fig. 12. Pulse sequences for Grover's algorithms without and with a warp-drive gate. The first row shows the pulse sequence which picks the file "10". The second row is the pulse sequence obtained by decomposing UcpU\o, where J/QP is the warp-drive gate.
The spectra shown in Fig. 13 demonstrate that the "warp-drive gate" works properly. This time, the pseudopure state was prepared by spatial labeling, which is discussed in § 4.3.2. Subsequently the gate [Tin (C^CP^IO) was applied on the pseudopure state |00). See the pulse sequence in the first (second) row in Fig. 5.4.4. The effectiveness of the "warp-drive" gate should be obvious from the fact that the signal of UcpUw is much sharper than that of Uio- Note that we should get |11) with UcpUio instead of 110). 6. DiVincenzo criteria Here we briefly discuss whether the room-temperature liquid-state NMR quantum computation fulfills the DiVincenzo criteria 4 : (1) Be a scalable physical system with well-defined qubits: The NMR quantum computation employs the spins of atomic nuclei as qubits. We need to prepare appropriate molecules that contain spin-1/2 nuclei. Typical nuclei are hydrogen, fluorine, carbon-13, and nitrogen-15. The scalability, however, may
181
6.0 ppm Ul0\00)
Fig. 13. Grover's database search algorithm with and without a warp-drive gate. The gates Uio and UcpUio are applied on the pseudopure state |00), where UCP is one of the warp-drive gates for U\Q.
not be so obvious. The maximum number of qubits manipulated with a room-temperature liquid-state NMR quantum computer 2 so far is seven. The difficulty in scalability stems from the lack of an efficient method of initialization as is discussed next. It should be also noted that selective addressing to each qubit requires sufficient chemical shifts in resonance frequencies, which becomes more and more difficult as the number of spin increases. (2) Be initializable to a simple fiducial state such as |000 . . . ) : Nuclear spins are subject to a strong magnetic field, typically on the order of 10 T, in thermal equilibrium at room temperature. The spins are in a highly mixed state far from a pure state required for quantum computations. One can prepare a pseudopure state, which effectively behaves like a pure state, by using temporal averaging, spatial averaging or logical labeling. These schemes, however, come with a cost; either the signal intensity decreases exponentially, or the number of required measurements grows exponentially as the number of qubits increases. Another important issue related to a pseudopure state, or
182
equivalently a mixed-state, is the lack of entanglement. 3 Optical pumping and various solid-state NMR schemes may fulfill this criterion, but they are beyond the scope of the present discussion. (3) Have much longer decoherence times, and (4) Have a universal set of quantum gates: The third and the fourth criteria are discussed together. Onequbit gates are realized with various rf pulses, whose duration is of the order of 10 fj,s, while the CNOT gate can be implemented with several rf pulses and time evolution by a J-coupling between two nuclei, which takes of the order of 10 ms in total. Therefore, the universal set of quantum gates is provided in NMR quantum computation. Nuclear spins in a molecule dissolved in a liquid have very long coherence times of more than a few seconds, when properly prepared. The number of feasible gate operations within decoherence time may be reasonably large to execute simple algorithms. (5) Permit high quantum efficiency, qubit-specific measurements: Readout of nuclear spin states is a well established technique in NMR. It should be remembered, however, that readout is an ensemble measurement of many nuclear spins, which may require a modification of certain quantum algorithms. 36 In summary, the room-temperature liquid-state NMR technique in its present form cannot be a true candidate for a scalable quantum computer unless a drastically new improvement is developed. However, its potential importance should not be underestimated. The difficulty arises from the fact that the spin polarization of molecules in a liquid at room temperature is very small, of the order of 10~ 5 . If one could have a highly polarized "molecules", NMR-inspired techniques would lead to a true quantum computer. These "molecules" are not necessarily natural but can be artificial. Electrons on surface of superfluid helium 37 can be a good candidate for such artificial molecules. We regard an NMR quantum computer as a good simulator for realistic quantum computers. We use an NMR quantum computer with several qubits in our research to investigate algorithm acceleration, decoherence simulation, decoherence suppression and other important issues in the physical realization of quantum computers. We strongly believe that the role
183 played by an N M R q u a n t u m computer is significant in the fundamental research in these fields.
Acknowledgements We would like t o t h a n k Toshie Minematsu for assistance in N M R operations and Katsuo Asakura and Naoyuki Fujii of J E O L for assistance in N M R pulse programming. MN would like to t h a n k partial supports of Grantin-Aids for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology, J a p a n , Grant No. 13135215 and from J a p a n Society for the Promotion of Science (JSPS), Grant No. 14540346. ST is partially supported by J S P S , Grant Nos. 15540277 and 17540372.
References 1. M. A. Nielsen, and I. L. Chuang, Quantum Computation and Quantum Information, (Cambridge University Press, Cambridge, 2000). 2. L. M. K. Vandersypen, M. Steffen, G. Breyta, C. S. Yannonl, M. H. Sherwood, and I. L. Chuang, Nature 414, 883 (2001). 3. S. L. Braunstein, C. M. Caves, R. Jozsa, N. Linden, S. Popescu, and R. Schack, Phys. Rev. Lett. 83, 1054 (1999). 4. D. P. DiVincenzo, http://www.research.ibm.com/ss_computing/. See, also, D. P. DiVincenzo, Science 270, 255 (1995). 5. D. G. Cory, A. F. Fahmy, and T. F. Havel, Proc. Natl. Acad. Sci. USA 94, 1634, March (1997). 6. J. A. Jones, Prog. NMR Spectrosc. 38, 325 (2001). 7. L. M. K. Vandersypen and I. L. Chuang, Rev. Mod. Phys. 76, 1037 (2004). 8. L. M. K. Vandersypen, Stanford University Thesis (2001). 9. For example, see T. E. W. Claridge, High-Resolution NMR techniques in Organic Chemistry, (Elsevier, Amsterdam, 2004) 10. R. R. Ernst, G. Bodenhausen, and A. Wokaun, Principles of Nuclear Magnetic Resonance in One and Two Dimensions, (Oxford University Press, Oxford, 1991). 11. F. Bloch and A. Siegert, Phys. Rev. 57, 522 (1940). 12. N. F. Ramsey, Phys. Rev. 100, 1191 (1955). 13. R. Laflamme, E. Knill, D. G. Cory, E. M. Fortunato, T. F. Havel, C. Miquel, R. Martinez, C. J. Negrevergne, G. Ortiz, M. A. Pravia, Y. Sharf, S. Sinha, R. Somma, and L. Viola, Los Alamos Science Number 27, 226 (2002). 14. D. G. Cory, R. LaBamme, E. Knill, L. Viola, T. F. Havel, N. Boulant, G. Boutis, E. Fortunato, S. Lloyd, R. Martinez, C. Negrevergne, M. Pravia, Y. Sharf, G. Teklemariam, Y. S. Weinstein, and W. H. Zurek, Fortschr. Phys. 40, 875 (2000), J. A. Jones, Fortschr. Phys. 40, 909 (2000). 15. H. De Raedt, K. Michielsen, A. Hams, O. Miyashita, and K. Saito, Eur. Phys. J. B27, 15 (2002).
184 16. A. Barenco, C.H. Bennett, R. Cleve, D.P. DiVincenzo, N. Margolus, P. Shor, T. Sleator, J. Smolin and H. Weinfurter, Phys. Rev. A 52, 3457 (1995). 17. M. A. Pravia, E. Fortunato, Y. Weinstein, M. D. Price, G. Teklemariam, R. J. Nelson, Y. Sharf, S. Somaroo, C. H. Tseng, T. F. Havel, D. G. Cory, Concepts Magn. Res. 11, 225 (1999). 18. U. Sakaguchi, H. Ozawa, and T. Fukumi, Phys. Rev. A 6 1 , 042313 (2000). 19. Y. Sharf, T. F. Havel, D. G. Cory, Phys. Rev. A 62, 052314 (2000). 20. N. A. Gershenfeld and I. L. Chuang, Science 275, 350 (1997), L. M. K. Vandersypen, C. S. Yannoni, M. H. Sherwood, I. L. Chuang, Phys. Rev. Lett. 83 (1999) 3085. 21. D. Deutsch and R. Jozsa, Proc. Roy. Soc. London, Ser. A bf 439 553 (1992). 22. http://www.jeol.com/nmr/nmr.html. 23. J. A. Jones, M. Mosca, and R. H. Hansen, J. Chem. Phys. 109, 1648 (1998). J. A. Jones and M. Mosca, Nature 393, 344 (1998). 24. I. L. Chuang, L. M. K. Vandersypen, X. Zhou, D. W. Leung, and S. Lloyd, Nature 393, 143 (1998). 25. N. Khaneja, R. Brockett, and S. J. Glaser, Phys. Rev. A 63, 032308 (2001). 26. M. Nakahara, J. J. Vartiainen, Y. Kondo, S. Tanimura, and K. Hata, Phys. Lett. A, to be published and eprint quant-ph/0411153. 27. Y. Makhlin, Quant. Info. Proc. 1, 243 (2002). 28. A. M. Childs, H. L. Haselgrove, and M. A. Nielsen, Phys. Rev. A 68, 052311 (2003). 29. J. Zhang, J. Vala, S. Sastry, and K. B. Whaley, Phys. Rev. A 67, 042313 (2003). 30. L. K. Grover, Phys. Rev. Lett. 79, 325 (1997). 31. E. Farhi and S. Gutmann, Phys. Rev. A 57, 2403 (1998). 32. I. L. Chuang, N. Gershenfeld, M. Kubnec, Phys. Rev. Lett. 80, 3408 (1998). 33. M. Nakahara, Y. Kondo, K. Hata, and S. Tanimura, Phys. Rev. A 70, 052319 (2004). 34. A. O. Niskanen, J. J. Vartiainen, and M. M. Salomaa, Phys. Rev. Lett. 90, 197901 (2003). 35. J. J. Vartiainen, A. O. Niskanen, M. Nakahara, and M. M. Salomaa, Phys. Rev. A 70, 012319 (2004). 36. N. A. Gershenfeld and I. L. Chuang, Science 275, 350 (1997). 37. P. M. Platzman, M. I. Dykman, Science 284 1967(1999).
185
Optical Quantum Computation Kae Nemoto National Institute of Informatics, 2-1-2 Hitotsubashi, Chiyoda-ku, Tokyo 101-8430, E-mail: [email protected]
Japan
W. J. Munro Hewlett-Packard Laboratories, Filton Road, Stoke Gifford, Bristol BS34 8QZ, UK E-mail: [email protected] T. P. Spiller Hewlett-Packard Laboratories, Filton Road, Stoke Gifford, Bristol BS34 8QZ, UK E-mail: [email protected] We review recent theoretical progress in finding ways to perform quantum processing with optics using photon encoded qubits, qudits and continuous variables. We discuss the requirements, advantages and disadvantages of each approach and show that optics holds significant promise for computation.
1. Introduction There are currently numerous possible routes forward for quantum computation and communication hardware 1,2 . A significant number of these are based on coherent condensed matter systems 3 , however another strong candidate (especially for communication) would be to use states of the electromagnetic field (photons). Optics has already played a major role in the testing of fundamental properties of quantum mechanics5 and, more recently, implementing simple quantum information protocols. This has been possible because photons are easily produced and manipulated and, as the electro-magnetic environment at optical frequencies can be regarded as vacuum, are relatively decoherence free. If we consider quantum information processing and computation using photons then we have a number of distinct techniques we can use to encode and process this information. These
186
include but are not limited to: • • • •
single photon linear optics single photon nonlinear optics qudit information processing continuous variable (CV) information processing (qunat information processing) • Hybrid systems using qubits, qudits and qunats.
Now that we have started to consider quantum information processing with light, we need to define a set of criteria to discuss to help us evaluate its potential. Essentially all aspects of quantum information processing can be reduced to three stages: prepare; evolve; measure. As we are classical objects, and can only relate to conventional information, we have to carefully prepare our quantum information (qubits, qudits or qunats) in appropriate states, allow them to evolve (feeding in the classical information pertaining to the process), and then measure them to extract information in a form we can use. A well known set of requirements that need to be satisfied so this information processing can be performed has been summarized in the universally accepted DiVincenzo Criteria 6 ' 7 : (1) A collection of well-characterized qubits, qudits or qunats is needed. One at a time will do for cryptography although entangled pairs could be a useful luxury. Controlled interactions between a few qubits, qudits, qunats are needed for small scale processing. Scalability of the number of qubits, qudits or qunats is necessary for full blown quantum computing. (2) Preparation of known initial states for the qubits, qudits or qunats must be possible. The purer the better. (3) The quantum coherence of the system(s) must be maintained to a high degree during the evolution/computation stage, giving a high fidelity for the final state. For few-qubit processing it may suffice to have a straight shot at the process with good qubits, qudits or qunats and gates; for large scale quantum computation error correction will almost certainly be needed 8 . For fault-tolerant operation the fidelity of individual gates probably needs to be 0.999 or better. (4) The unitary quantum evolution required by the algorithm or protocol must be realizable. Some unitary processes are much easier in experiment than others. As with conventional computing, the minimum of a universal set of elementary gates must be possible.
187
(5) High fidelity quantum measurements on specific qubits, qudits or qunats must be possible, in order to read out the result of the process or computation. (6) The capability to interconvert stationary (processing or memory) and flying (communication) quantum information must exist. (7) It must be possible to transmit flying qubits, qudits or qunats coherently between specified locations. For a localized quantum processing and computing device the first five criteria (l)-(5) needed to be satisfied, with the last two (6) and(7) required for genuinely spatially separated processing. Criterion (4) indicates that the unitary quantum evolution required by the algorithm or protocol must be realizable. However there may be certain algorithms that do not need a universal set of quantum gates to implement them. The potential problem with this is that without the use of a universal set of quantum gates, the quantum process may be able to be efficiently classically simulated. Just because our quantum processes create entangled states does not mean that this process can not be efficiently classically simulated. The concept of what can be efficiently classically simulated is summarized in the GottesmanKnill (GK) theorem 9 ' 10 which we will discuss below.
1.1. The Gottesman-Knill
theorem
It is clear that certain quantum information processing tasks can be implemented much more easily than others. Simple schemes in quantum communication require only superposition for the quantum schemes to be distinguished from their corresponding classical ones. However, for computational tasks, simply having superpositions of quantum states is not enough to achieve truly quantum computation, distinguished from classical computation. We need to consider the concept of universality. Universality is an important concept that allows quantum computational circuits to be distinguished from their classical counterparts. The GK theorem 9 ' 10 states that any quantum algorithm that initiates in the computational basis and employs only a restricted class of gates (Hadamard, phase, CNOT, and Pauli gates), along with projective measurements in the computational basis, can be efficiently simulated on a classical computer. This means there is no computational advantage in using such circuits on a quantum computer. A classical machine could simulate the circuit efficiently. This set of gates described above preserves the Pauli group, which consists of Pauli operators and the identity operator / with
188
coefficients of { ± l , ± i } . For instance, the Pauli group for one qubit is { ± c x , ±cry, ±crz, ±1, ±i
Z(p) = eiri,
(1)
with q,p £ R. Based on the generalized Pauli operators, we can now find operations which preserve the generalized Pauli operators. This can be done by generalizing the gate set of the GK theorem for qubits. The Fourier transform F is the qunat analog of the Hadamard transformation. It is defined as F = exp
l-A?+?) ,
U4 and the action on the Pauli operators is
(2)
FX{q)F* -> Z(q) FZ(p)Fi
- X{p)~l.
(3)
The "phase gate" P{rj) for qunats is a squeezing operation, defined by P{r}) — exp
1
-i
(4)
189
and the action on the Pauli operators is P(T,)X{q)P(V?
-
e^X(q)Z(Vq),
Pfa)Z(p)Pfo)t -> Z{p).
(5)
(The operator P{rj) is called the phase gate, in analogy to the discretevariable phase gate P, because of its similar action on the Pauli operators.) The two-bit operation, CNOT gate, can be generalized to the SUM gate as SUM = e x p ( - ^ i ® p 2 ) .
(6)
The action of this gate on the Pauli group for two systems is given by SUMXi(g)
(7)
Now, the generalized GK theorem for qunats can be summarized as follows: Any qunat quantum information process that begins with Gaussian states (products of squeezed displaced vacuum states) and performs only • • • •
linear phase-space displacements (given by the Pauli group), squeezing transformations on a single oscillator system, SUM gates, measurements in the position- or momentum-eigenstate basis (measurements of Pauli group operators) with finite errors and losses, • classical feedforward, can be efficiently simulated using a classical computer. The conditions can be simplified by transformations generated by inhomogeneous quadratic Hamiltonians in terms of the canonical operators {q%, pi; i = l , . . . , n } . Thus, any circuit built up of components described by one- or two-mode quadratic Hamiltonians [such as the set of gates SUM, F, P{rj), and X(q)}, that begins with finitely squeezed states and involves only measurements of canonical variables may be efficiently classically simulated. 1.2. Universal
Quantum
Computation
Implementing a universal set of gates is certainly a way to achieve universal quantum computation. However, in real quantum mechanical systems
190
it is generally difficult to implement such a universal gate set. Generally speaking, systems with small decoherence have difficulties with controlled operations between two (or more) qubits, and where such controlled operation are natural, it is often difficult to maintain quantum coherence in the system and to ensure the accessibility to each individual qubit. Given the fact that universal quantum computation requires both good quantum coherence and entanglement in the computational system, this trade-off relation between quantum coherence and multi-qubit interaction seems to be an obstacle to universal quantum computation, when one considers the requirement for universal gate sets. However, implementation of universal gate sets is not the only way to achieve universal quantum computation. In fact, some implementations of non-universal gate sets can achieve universal quantum computation in a scalable manner. Such implementations are not allowed to access to the entire Hilbert space, yet it is possible for these gates to simulate universal quantum computation on a lower dimensional subspace. For qubits restricted to states with just real amplitudes, which is called a rebit subspace, the controlled-rotation gate is such an example 12 . Universal quantum computation can be achieved through various approaches. In addition to gate-based schemes, universal quantum computation can also be achieved by measurement alone, or through schemes based on measurements 13-15 . In classical computation, measurement schemes are trivial, and hence universal classical computation is considered just in terms of gate-based schemes. However, in quantum computation this is not the case. Measurement alone can be as powerful as a universal gate set. A number of measurement-based universal quantum computation schemes have been proposed. Universal quantum computation can be broadly classified in three categories 16 : • Universal set: A universal set can construct an arbitrary circuit with an arbitrary precision. • Universal computational set: A universal computational set is not universal by the definition of universality, but can simulate universal computation with polynomial extra resource. • Non-universal set: A non-universal set can be efficiently simulated on a classical computer and cannot simulate universal computation with polynomial extra resource. It can construct universal computation only with additional measurement strategies. Now because of the power of measurement in quantum computation, it is obvious that it does not make much sense to evaluate quantum compu-
191
tation schemes solely either on gate sets or on measurement schemes. This aspect of quantum computation can be exploited to construct a circuit bypassing the difficulties of some particular operations. However, at the same time it adds an extra complication into our criteria for universal quantum computation. 1.3. Back to the DiVincenzo
criteria
Now let us return to our discussion of the DiVincenzo criteria. The demands from these seven criteria are extremely tough indeed and the construction of small but useful quantum processors will certainly not be easy. Nevertheless, the goal of performing quantum simulations beyond what we can do with any conventional computer seems to be a good challenge for the next decade of research. If we can get that far, it will be rather easier to make predictions about the bigger goal of large scale many-qubit computing. This article will be structured into four main sections, the first being on linear optical quantum computation, the second being on continuous variable quantum computation and the third on hybrid schemes that combine both single photons and continuous variables. This article will finish with a brief discussion of recent experiments and the prospects for optical quantum information processing. 2. Single-photon quantum computation: processing using only linear optics Now let us begin our discussion of quantum computing with single photons. As we mentioned in our introduction, optics with single photons seems to be a natural candidate for quantum computation and information processing. For decades, the photon has been studied extensively by theorists and experimentalists alike and its properties are well known. Polarization information can be encoded on a single photon easily and this leads to an excellent qubit. For a polarization-encoded qubit an arbitrary state can be represented as cos6>|#)+e i0 sin<9|VO
(8)
where \H) and |V) represent the horizontal and vertical polarization of that single photon and are the basis states for the qubit. The general state is parametrized by two angles 9 and 0. This space can be pictured as the surface of a sphere, as illustrated in Figure 1. One of the greatest advantages of encoding in polarization is that single qubit operations are deterministic and trivial. They are simply achieved with linear elements such as
192
|H>-|V> |H>+i|V?
H>-i|V> H)+|V> An arbitrary state
Fig. 1. T h e state space of a polarisation-encoded qubit. Classical bits only have access to the two polar states, but an arbitrary qubit state can be anywhere on the surface of the sphere. Examples of particular (unnormalized) equatorial states are labeled.
beam-splitters, polarizing beam-splitters and phase shifters. Unfortunately, an entangling two-photon gate such as the CNOT gate is far less straightforward, because, in ordinary materials, photons do not interact with each other. The lack of nonlinearity had been a fatal problem in the optical implementation of a universal quantum computer until a recent proposal by Knill, Lafiamme and Milburn 17 . This linear optics quantum computation scheme exploits a new non-deterministic process to generate nonlinearity only from linear elements, single-photon sources and photon number detection. These non-deterministic gates can then be teleported into the computational circuit allowing their operation to be deterministic in terms of the computation. We will now describe this protocol in detail. 2.1. Linear optical quantum
computation
Recently, Knill, Lafiamme and Milburn (KLM) 17 proposed an elegant scheme to sidestep the requirement for strong nonlinear materials. They showed that it is possible to generate an effective large nonlinear interaction between single photons using passive optical elements and photo-detection. In this scheme the optical mode of interest is combined with several other ancilla modes by passive linear optical circuits (beam-splitters) and conditioned on obtaining particular detector signatures in the output ancilla modes. The process is non-deterministic, but successful operation is flagged by the ancilla measurement result, that is to say, successful operation is heralded.
193
The basic operation they considered was a nonlinear sign shift on a general two-photon state, a state with two or less photons. This conditional operation transform the two-photon state according to c0|0) +ci|l> +c 2 |2) -> c0|0) +ci|l> - c 2 | 2 )
(9)
where \n) is the nth Fock state of the optical field and the coefficients Cj are complex amplitudes which satisfy the usual normalization constraint ^ \CJ\2 = 1. The implementation of this gate with linear optical elements and single photon detection is shown in Figure 2.
|\m
.
I
^^
'
^IW) /
v' 'V
|i) |0)
Conditioned output state
/fc^-1-^
^BS\
<
'
2
/—#-|i>
^BSV-^lo) Photon number resolving detector
Fig. 2. Schematic diagram of the KLM nonlinear sign shift gate. This gate is composed of three input states, three beam-splitters and two photon number resolving detectors. The three input states are our unknown signal state I*) = co|0) + c\|1) + C2|2), plus two ancilla states, a single photon |1) and a vacuum state |0) respectively. These three states interact with each other via the beam-splitters characterized by 6\, 62 and 9\ (where cos 2 6\ = * ~ and cos 2 62 = 3 — 2\/2). After the interaction of the three states with the beam-splitters, the ancilla states are detected using single photon number resolving detectors. When a photon is detected in the first ancilla and the vacuum in the second ancilla, the signal state | * ) is transformed to |\l>') = co|0) + c i | l ) — C2|2). We observe the sign of the |2) amplitude has flipped and hence the name of the gate being a nonlinear sign shift.
Let us now consider the action of this gate in further detail. Consider that our signal state is the vacuum |0). After the interaction with the beamsplitters and the detection of one photon in the first ancilla and the vacuum in the second ancilla, the signal state is transformed according to |0) -> [cos2 0, cos 02 + sin 2 0X] |0) .
(10)
Similarly the one-photon signal state |1) is transformed according to |1) -> - [cos2 6>i cos2<92 + sin2 Bx cos<92] |1> •
(11)
Lastly the two-photon signal state |2) is transformed according to |2) - > - c o s 0 2
'1 - cos 0 1 { 3 c o s 2 < 9 2 - l } + s i n ' ! 0 1 COS #2 |2> .
(12)
194
Now if we choose cos2 8\ = |0>-i|0),
4
* /» and cos2 #2 = 3 — 2\/2, then |1>->||1>,
|2>->~|2>
(13)
We see immediately that all three transformations have the same \ coefficient with the 12) component also having a negative sign. The A squared is the probability that the transformation occurs. This means that the general state \%f) will be transformed according to | * ) = c 0 | 0 ) + c 1 | l ) + c 2 | 2 ) ^ i | * ' ) = ^[co|0)+c1|l)-c2|2)] .
(14)
The loss in amplitude reflects other outcomes, and so the transformation (9) is effected with a success probability of 1/4. This probabilistic but heralded nonlinear sign shift gate is the core gate in the KLM linear optical quantum computation scheme. From this gate the usual CNOT gate can be constructed. Before we indicate how to construct the CNOT gate, which with single qubit gates forms a universal set of gates, we will examine several variants for the construction of nonlinear sign shift gates 18 ' 19 . These gates are depicted in Figure 3. Both of these variant gates enable the sign to be flipped on the 12) amplitude but require one fewer beam-splitter. There is a slight cost to the efficiency of this variant gates due to having one less beamsplitter. The probability of success reduces from 25 percent to roughly 23 percent 18 . Now let's consider the construction of a CNOT gate from these nonlinear sign shift gates. The design is rather simple and is depicted in Figure 4. As a controlled-phase (CPhase) gate is equivalent to the CNOT gate up to local operations, we will consider this form first. The CPhase gate is composed of three basic elements, a 50/50 beam-splitter followed by two nonlinear sign shift gates and finally another 50/50 beam-splitter. Let us now examine how this gate acts on a general two qubit state (where our qubits have been encoded in photon number) /30|00> + /?i|01) + /32jl0> + /33|11). The first qubit state represents the control qubit and the second the target. After the first 50/50 beam-splitter this state is converted (transformed) to A)|00> + A
[|01> - |10>] + A
[|10> + |01)] + A
[|02) - |20>] .
(15)
Then applying separate nonlinear sign gates to both modes we get A)|00) + A
[|oi> - |io>] + A
[|i 0) + | 0 i } ] _ i | [j02> - |20>]
(16)
195
a) -
|1>
e2
^
|o>
5^
|i>
9i
-#-|o>
b) |o>-
|o>
->v X
^4
-/RSN 10 2
•ID
ID'
Fig. 3. Schematic diagram of the two variants of the KLM nonlinear sign shift gate. Both of these gates are composed of three input states, two beam-splitters and two photon number resolving detectors. Again the three input states are our unknown signal state I*) = co|0) + ci11) + ci\l) plus the ancilla states |1) and |0) respectively. These three states interact with each other via the beam-splitters characterized by 9\ and 62 (where cos 2 0\ — 3 ~ 7 and cos 2 02 = 5 — 3\/2). When a photon is detected in the first ancilla and the vacuum in the second ancilla, the signal state |*) is transformed to I*') = co|0) + c i | l ) - C2|2). We observe that the sign of the |2) has flipped.
y
/•-
W
SW/ii)^.
RS
Target in
NSgmc-v
"TTS? VJO/JON
r
BS
Target out
Fig. 4. Schematic diagram of a probabilistic but heralded linear optical controlled-phase (CPhase) gate. The CPhase gate is composed of two nonlinear sign shift gates (NS-gates) plus two 50/50 beam-splitters.
where we observe that the sign of the fa component has nipped due to the
196
action of these gate. Finally applying the last 50/50 beam-splitter we have A,|00)+/?i|01)+/? 2 |10)-/?3|ll}
(17)
which is what we would expect if a CPhase gate had been applied to our initial state. We must re-emphasize that this is not a deterministic gate. It is probabilistic but heralded in nature due to the two nonlinear sign shift gates used to construct it. The CPhase gate has a 1/16 probability of success. In the above discussions on nonlinear sign shift and CPhase gates we have considered photon number states of light and not polarization states such as \H) and |V). It is well known that we can convert the latter to photon-number encoded states by a polarizing beam-splitter. Basically \H) - • |10)
(18)
\V) -
(19)
|01)
We call this a dual rail qubit since the single mode polarization state has been converted to a which-path type qubit. This dual rail qubit is essential because single qubit operations can be performed using only beam-splitters and phase shifter. This is not the case for a qubit encoded in the basis states |0) and |1). Control out
Control in
Target in \ J ^
7
l 50/50 BS
50/50 BS
"1"°"™
In ., •
I 50/50 BS
5 0 / 5 0 B S E = ^ 3 Target out
Fig. 5. Schematic diagram of a probabilistic but heralded linear optical CNOT gate. The CNOT gate is composed of two dual rail qubits (each qubit has the basis states 110) and |01)). The qubit is effectively encoded in which-path information. The CNOT operation begins by performing a Hadamard operation on the target qubit utilizing a 50/50 beam-splitter. Then the previous CPhase gate is applied to one mode from the control qubit and one mode from the target qubit. The CNOT operation is then completed by implementing a further Hadamard operation on the target qubit.
This dual rail encoding then allows us a natural way to transform the CPhase gate to a CNOT gate (depicted schematically in Figure 5). The CNOT gate is basically the CPhase gate with a Hadamard operation being performed before and after it on the target qubit. So far we have shown how it is possible to construct heralded but probabilistic two-qubit gates in linear optics. The CPhase and CNOT gates have
197
a maximum probability of success of 1/16. A natural question is whether we can increase this probability of success. Recently Knill 20 has shown a slightly different CNOT circuit (not composed of nonlinear sign shift gates) which has an ideal probability of success of 2/27. Even with classical feedforward it is not possible to significantly increase this. However, if we allow the use of entangled resources the probability of success for a CNOT gate for instance can be increased to one quarter 21 . The circuit is depicted in Figure 6. Control out
Target out
Control in
Target in
Fig. 6. Schematic diagram of Franson's four-photon CNOT gate. It is composed of two single-polarization encoded qubits (the control and target) plus one maximally entangled Bell state of the form \HH) + | VV), plus polarizing beam-splitters (operating either in the {H, V} or {D = H + V, D = H — V} bases) and two photon number resolving detectors. When the ancilla photons are detected at the LHS and RHS detectors, a CNOT operation is induced onto the control and target qubits.
Let us consider the operation of this gate in more detail. We will assume that the control and target are initially in the state co\HH)ct + cx\HV)ct
+ c2\VH)ct
+ c3\VV)ct
(20)
with the ancilla qubits prepared in the maximally entangled state \HH)ab + \VV)ab. This four-qubit state can be written as co\HHHH)cabt
+ c0\HVVH)cabt
+
Cl\HHHV)cabt
+ Cl\HVVV)cabt
+
c2\VHHH)cabt
+ c2\VVVH)cabt
+
cz\VHHV)cabt
+ c3\VVVV)cabt
.
(21)
The action of the LHS {H,V} polarizing beam-splitter between the control and ancilla mode a bunches the photons if the polarizations are different
198
and anti-bunches them if they are same. This results in the four-mode state co\H, H, H, H)cabt
+ c0\HV, 0, V, H)cabt
+
d \H, H, H, V)cabt
+
+
Cl
\H V, 0, V, V)cabt
c2\0, VH, H, H)cabt
+ c2\V, V, V, H)cabt
+
c 3 |0, VH, H, V)cabt
+ C3\V, V, V, V)cabt
(22)
where the commas have been inserted to make it explicit which photons are in which mode. As a part of our protocol, we are going to condition the control and target qubits evolution on a single photon being detected in mode a. The protocol is aborted if zero or two photons are detected. Hence the above state conditioned by a single-photon detection in mode a is c01H, H, H, H)cabt
+ cx | H, H, H, V) cabt
+ c2\V, V, V, H)cabt + c3\V, V, V, V)cabt . (23) For the detection process we measure the photon polarization in the {D, D} basis. For a click at the D detector in mode a we have c0\H, H, H)cbt + Cl\H, H,V)cbt
+ c2\V,V,H)cbt
+ c3\V,V,V)cbt
(24)
while for a D click we have c0\H, H, H)cbt + a\H, H, V)cbt - c2\V, V, H)cbt - c3\V, V, V)cbt . (25) A sign flip (via classical feedforward) on the V polarization of the control qubits transforms the state (25) to (24). There is only a 50 percent chance that we will detect a single photon in ancilla a. The second part of the protocol is now to have the second ancilla qubit b and the target qubit interact on a polarizing beam-splitter operating in the {D, D} basis. Before we do this it is useful to rewrite the ancilla b and target qubit of (25). The state (25) then looks like l(co + ci)\H)c
+
{c2+c3)\V)c]\DD)bt
+ [(co - ci) \H)C - (c2 - c3) \V)C] \DD)bt + [(co - ci) \H)C + (c2 - c3) \V)e] \DD)bt + [(co +
Cl)
(26)
\H)C - (c2 + c3) \V)C] \DD)bt .
In this form we can see that the \DD)bt and \DD)bt have the same polarization and hence will not bunch on the polarizing beam-splitter. The terms \DD)bt and \DD)bt are of opposite polarization and do bunch. If we require that one photon is detected in mode b then only the terms l(c0 + Cl)\H)c
+
(c2+c3)\V)c}\DD)bt
+ [(c0 - ci) \H)C - (c2 - c3) |V)C] \DD)bt
(27)
199
can contribute. Now performing a measurement of the ancilla mode in the {H, V} basis, our control and target qubits are transformed to c0\HH)ct + Cl\HV)ct
+ c2\VV)ct + c3\VH)ct
(28)
+ c2\VH)ct + c3\VV)ct
(29)
for an H result, and c0\HV)ct + Cl\HH)ct
for a V result. The second case can be transformed to the first by a bit flip of target qubit. The state (28) is what we would expect if a CNOT operation was performed on the initial state (20). This CNOT operation is probabilistic (but heralded) and works 25 percent of the time. Unfortunately these probabilistic gates cannot be used directly for scalable quantum computation because the probabilistic nature would lead to an exponential decrease in the probability of the entire computation being successful. 2.2. Qubit
Teleportation
A very elegant solution to this problem has been proposed by Gottesman and Chuang 22 who have shown that quantum teleportation 23 can be used as a universal quantum primitive. In essence, quantum teleportation allows for the fault-tolerant implementation of "difficult" quantum gates that would otherwise corrupt the fragile information of a quantum state. The Gottesman and Chuang gate teleportation scheme22 is depicted in Figure 7. Let us consider this teleportation primitive in some detail. Suppose that we want to apply a CNOT operation between a control and target qubit. We assume that our control and target are initially in the state (20). We begin by creating two Bell states of the form \HH) + \VV). We write their product in the form \HHHH)abef
+ \HHVV)abef
+ \VVHH)abef
+ \VVVV)abef
. (30)
A CNOT operation is performed between the b and e qubits. When the gate is successful this operation creates the four qubit entangled state \HHHH)abef
+ \HHVV)abef
+ \VVVH)abef
+ \VVHV)abef
. (31)
Now consider that a Bell measurement is performed between the control qubit and ancilla qubit a and between the target qubit and ancilla qubit /. From each Bell measurement we get one of the four results \HH) + \VV), \HV) + \VH), \HH) - \VV) or \HV) - \VH). The last three Bell states are related to the first by an X, Z or XZ operation. Hence we can specify the Bell-state measurement outcomes as J, X, Z or XZ. Consider that both
200
Control in
Bell Measurement
Bell state {
ICNOT Bell state
Control out Target out
Target m
measurement dependent ^mmr classical feedforward Bell Measurement
Fig. 7. Schematic diagram illustrating the teleportation of a two qubit gate. The scheme is composed of our initial control and target qubit plus two ancilla maximally entangled Bell states. The Bell states are entangled by the CNOT operation. This operation does not need to be deterministic, having a heralded result is sufficient. Only after the CNOT operation is successful, a Bell measurement is performed between the control and one of the qubits of the first Bell state and another between the target qubit and one of the qubits of the second Bell state. Classical feed forward operations perform bit/phase flips on the teleported qubits depending on the result of these measurements (as detailed below). This generates a CNOT operation between the control and target qubits.
Bell-state measurements give us the result I (that is, the \HH) + \VV) state is detected). In this case the ancilla qubits b and e are transformed to co\HH)be + Cl\HV)be
+ c2\VH)be + c3\VV)be
(32)
which is the state we would obtain had we applied the CNOT gate directly to the initial state (20). Similarly if the result of the Bell state measurements is X for the first measurement and / for the second we would obtain the state co\VH)be + Cl\VV)be
+ c2\HH)be + c3\HV)be
(33)
which can be transformed to (32) by applying a bit flip (an X operation) to the first qubit 6. Figure 7 indicates the classical feedforward operations needed to transform the states from all the Bell measurement results to (32). This really shows the power of gate teleportation. As long as we are able to perform Bell measurements efficiently, it is possible to transform probabilistic but heralded gates into deterministic ones. The key is that we keep applying the probabilistic gate to a known set of entangled states
201
until it works. This creates a new entangled state. We then teleport the computational qubits using this new entangled state, and the computational qubits we get out have the appropriate gate operation applied to them. This is a very powerful technique and is in fact a universal quantum primitive. Unfortunately, using only linear optics it is impossible to implement a Bell state measurement that distinguishes all four Bell states. With a single beam-splitter only two of the Bell states can be distinguished, and the scheme effectively works only 50 percent of the time. With two Bell-state measurements required for teleporting the CNOT gate the probability of success would be 1/4. To increase the success probability further we need to change the teleportation procedure 17 . Consider an initial entangled resource of the form n
\Tn) = ^2\iy\o)n-j\oy\i)n-j,
(34)
3=0
with |a)J' = |a)i <8> \a)^ ® ...® \a)j. This state \Tn) encodes n qubits with the fc'th qubit located in the modes k and n + k. The teleportation protocol using the state \Tn) teleports mode 0 in a superposition of |0) and |1) to one of the last n modes of our entangled resource \Tn). This occurs by a Bell measurement implemented using an n + 1 point Fourier transform implementable with passive linear optics and measurements of the number of photons in each of the modes 0 . . . n on the mode to be teleported and the first n modes of \Tn). If the result of the measurement is n = j then the teleported state is in the mode n + j . If the measurement result is 0 or n + 1, the Bell state measurement fails. This failure occurs with a probability l / ( n + 1) and so the probability of success for the teleportation is now n/{n + 1). For n = 1 we get our well known P = 1/2. However as n increases significantly this probability approaches unity. Given that we now have an efficient teleportation protocol let us return to the teleportation of gates. In this instance we will consider the CPhase gate instead of the CNOT. Consider now two copies of the state \Tn). These can be written in the form n
n
\Tn)\Tn) = ^ ^ | l ) ^ | 0 ) " ^ | 0 } ^ | l ) n ^ | l ) f e | 0 ) " - f c | 0 > f c | l ) " - f e
(35)
3=0 k=0
If we wish to teleport a CPhase gate then we need to apply a CPhase operation to each pair of modes (n + j , 3n + k) with j and k in the range 1 . . . n. When this operation is successful (remembering that the CPhase is
202
heralded) we obtain the highly entangled state n
|T^> = J2 (-l) (n_j)(n " fc) |l} j |0) n ^|0) i |l) n ^|l) fc |0) Tl - fc |0) fc |l} n - fc
(36)
The teleportation protocols are applied between the control mode and modes 1 . . . n and between the target mode and modes 3n + 1... An. Now given that we get measurements results (j, k) until sign flip corrections, the CPhase gate has been teleported on the modes (n+j, 3n+k). The probability of success now scales as n2/(n + l ) 2 due to the two Bell measurements. With this teleportation protocol we now have a mechanism for teleporting gates with near unit probability of success and hence we can achieve a universal set of gates (deterministic single qubits operation and the teleported CPhase or CNOT). The only disadvantage to this scheme is that the resource cost in generating the entangled state |T£) is high and potentially requires a very large number of single photons. This resource cost is polynomial and hence efficient. It does however raise the question of whether there is a more resource-efficient mechanism for single-photon linear optical quantum computation.
2.3. One-way
quantum
computation
An alternative approach to the usual gate based quantum computation schemes is the so-called cluster state or one-way quantum computation schemes. Using single photons, the cluster-state linear optical approach has a number of significant advantages including that it removes the need for the teleportation of non-deterministic gates 24 ' 25 . Hence this is much more efficient in terms of physical resources. There have been several proposals to generate linear optical cluster states. The technique we will discuss here was first proposed by Browne and Rudolph 25 . The core elements in the Browne and Rudolph cluster state scheme are two "fusion" mechanisms. These allow for the construction of entangled photonic states. The core resources required are a source of maximally entangled two qubit photonic states. These can either be constructed with the linear optical techniques described in the previous subsection (we only need a heralded result) or via a two photon entangled source. Given the Bell states, we can proceed to build up the cluster states using only non-deterministic parity-check measurements, which involve combining the photons on a polarizing beam-splitter (PBS) followed immediately by measurement on the output modes.
203
a)
b)
Single Photon Detector
Single P h o t o n Detector
4 5 Rotator
4 5 Rotator
Type I Fusion
Type II Fusion
Fig. 8. Schematic diagram of two non-deterministic"qubit fusion" gates. Both gates are based on polarizing beam splitters (PBS). The Type I gate (subfigure (a)) consists of combining two spatial modes on a PBS, and measuring one of the output modes with a polarization discriminating photon counter after a 45° polarization rotation. The TypeII fusion gate (subfigure (b)) is constructed from the Type I gate by adding both 45° rotations to each input mode and measuring the output modes in the rotated basis.
Browne and Rudolph introduced two fusion gates depicted in Figure 8. Both of these gates are based on Franson parity-check operation. Let us consider the action of the Type I "qubit fusion" gate. This operation by combining our two modes of interest on a polarizing beam splitter (PBS) and then rotating one of the output modes by 45° before measuring it with a polarization discriminating photon counter. Consider two Bell states of the form \HH) + \VV). We can write these in the form \HH)12\HH)34
+ \HH)12\VV)3i
+ \VV)l2\HH)3i
+
\VV)12\VV)3i(37)
Now taking the modes 2 and 3 and combining them on the PBS we get \HH)12\HH)34
+ \VV)12\VV)u
(38)
where we have kept only the terms in which one photon remains in each mode (this occurs 50 percent of the time). Now measuring in the {D,D} basis of the second mode we get upon the detection of one and only one photon \HHH)l3A
+ \VVV) 134
(39)
204
for the D result and \HHH)134
- |VW)134
(40)
for D. It is simple to convert the case (40) to (39), a maximally entangled GHZ state. Thus with 50 percent probability we can create a GHZ state from two Bell states. To generate three-qubit clusters (our core building block), we could just apply local operations to the GHZ state. Alternatively we can consider Bell states of the form \HH) + \VH) + \HV) - \VV). After the Type I fusion operation succeeds we obtain the state \HHH) + \HHV) + \VHH) + \VHV) + \HVH) - \HVV) - \VVH) + \VVV)
(41)
If this Type-I fusion gate is applied to the end-qubits of linear clusters (Figure 9(a)) of lengths n and m, a successful operation generates a linear cluster of length (n + m— 1). However for the Type-I fusion gate to operate successfully we need photon number resolving detectors. However, once we have generated the three-qubit clusters, we can join them via the Type-II fusion gate. Each time, with probability 1/2, the cluster grows in length by 2 qubits, or, equally likely, loses a qubit. A failed attempt creates a Bell pair from the 3-qubit cluster, which can be re-used for the generation of further 3-qubit clusters. Hence, on average, the cluster grows by 1/2 a qubit in length for each attempt, and the resources needed scale as (2 x 4 — 1) = 7 Bell pairs per qubit in the linear cluster. With more efficient strategies this resource cost can be decreased slightly further. It is however well known that one-dimensional cluster states are not sufficient for universal quantum computation. It is necessary as a minimum to create two-dimensional cluster states. This can be achieved using the Type-II fusion gate between the mid points of the linear chains (see Figure 9(b)). Thus it is possible to create arbitrary two dimensional cluster states and as arbitrary single-qubit measurements are easy to perform on photonic qubits we have an efficient mechanism to perform universal quantum computation. 2.4. Nonlinear
optics
Single photon nonlinear optics is very similar to the linear optical schemes described above, but uses other resources (rather than single photons, beam-splitters and single photon sources) to generate the nonlinear effects on the optical modes. In the linear optics schemes the nonlinear sign shift gate and CNOT (or equivalently CZ) gates were performed in a heralded
205
a)
b)
Type I Fusion
r^X.
Type Fusion
Fig. 9. Schematic diagram showing the fusing of two cluster states. In the first situation a) we fuse the end qubits of two linear clusters of length n and m. This creates a cluster of length (n + m — 1). In the second case b) the mid qubits of two linear clusters are fused to create a two-dimensional cluster with a cross-like layout. In this way non-trivial cluster layouts can be created.
but probabilistic fashion. This means that the teleportation trick was required to make the gates deterministic and hence significantly increased the required resources. It has long been known that a strong cross-Kerr nonlinearity is sufficient to create the CNOT gate (CZ, in fact, which may be converted to CNOT). However, natural materials do not currently process the massive, reversible nonlinearities required. Nevertheless, it is possible to engineer atomic systems to create large effective optical nonlinearities and it is these we will discuss briefly in this section. There are two regimes which are particularly important in using atomic systems to generate optical nonlinearities: • Passive optical nonlinearities 26,27 , • Near deterministic heralded optical nonlinearities 28 . Let us begin by examining the passive optical nonlinearities created by atomic systems. By this we mean nonlinearities generated without the need for active measurement. Electromagnetically induced transparency (EIT) which has recently been demonstrated in a number of situations is an ideal candidate for realizing controlled phase gates using the giant Kerr nonlinearities available 26 ' 27 . If we consider the four level (N) configuration depicted in Figure 10 below, then we can send the control mode into mode a and the target mode into mode c. The qubits could be encoded into whichpath or dual rail photons. In the latter case, the polarization encoding can
206
be turned into dual rail encoding using polarizing beam-splitters, as shown in Figure 10. Mode b is a classical pump, which is needed to link modes a and c. Such an EIT system generates an effective nonlinearity of the form U = e"ixtAaflc
(42)
where x is the strength of the nonlinearity and t is the interaction time. Only when there is a photon present in both the a and c modes passing through the atoms (the |11) amplitude) does U give a phase shift, that is U
(coo|0Q) + coi|01) + cio|10) + c i i | l l » = (coo|00) + coi|01) + cio|10) + c i i e - i x t | l l ) ) .
(43)
By choosing x< appropriately we can tune such that e~%x% = - 1 and hence generate a controlled phase shift of TT. This controlled phase shift together with local qubit operations is sufficient to generate a CNOT gate.
Fig. 10. Atomic systems (with a N configuration of levels) can be used to produce a CZ gate. T h e atoms provide an effective cross-Kerr nonlinearity, so passage of photons through the two rails that interact with the atoms (the |11> amplitude) will realize a phase shift t h a t can be adjusted to TT, to give a sign change relative to the other amplitudes. If the photons are initially encoded in polarization, this can be converted into which-path or dual rail encoding through polarizing beam-splitters. These send one polarization amplitude, representing |0>, around a path avoiding the atoms, and the other, representing |1), through the region containing the atoms.
207
Next let us consider the generation of heralded optical nonlinearities in a near deterministic fashion. The real key here is to use one of the key advantages from the all linear optical schemes, that is the conditioning measurement. However in these atomic cases we want the conditioning to be near-deterministic (with a success probability near to unity). An elegant scheme has been proposed by Gilchrist et al. 28 , which allows a nonlinear sign shift (NS) gate to be implemented with a probability exceeding 99 percent. Two of the NS gates can be used to build a CNOT gate. For this situation the conditioning measurement is done via a highly efficient iontrap shelving measurement and it is the efficiency of this measurement that makes the overall gate near-deterministic. As the gate is near-deterministic it is possible that the teleportation trick will not be needed and hence the overall resources required for the information processing will be less. This is the advantage of using heralded measurements to induce optical nonlinearities. So far we have considered a number of mechanisms to perform universal single-photon quantum computation. However as we have mentioned we can use other states of light, not restricted to qubits. We could perform qudit computation, but more importantly continuous variable computation is also available and this is what we will consider next. 3. Continuous variables As we have discussed, a qubit lives in a two-dimensional Hilbert space. However, some quantum systems live in larger spaces—even infinite-dimensional Hilbert spaces, such as for a particle moving in one dimension. Such a particle can be characterized by its position X and momentum P. In the context of quantum information, such a system is termed a qunat - the variables X and P have a continuous spectrum. Clearly, then, continuous variables are not restricted to the realm of optics. However, to date, the only continuous variables that have been successfully employed in quantum information are optical. Here X and P represent the two quadratures of an electromagnetic field mode (or the electric and magnetic fields). Given such a significant change in the way we encode our information, we need to carefully examine what operations are possible in these continuous variables systems and whether they could be universal. At first it may seem that quantum computation over continuous variables is an ill-defined concept. What does it mean to perform computation with continuous variables? First, there are an infinite number of parameters (each for a single particle or mode). Do we need to manipulate them all?
208
These are not easy questions to answer, but significant insight can be achieved by reconsidering universal computation with qubits. A qubit quantum computer could be defined as a device that applies local operations via quantum logic gates that affect only a few variables of this device at a time. By repeated application of these logic gates (for instance Hadamard, 7r/8 and the CNOT) we can effect any unitary transformation over a finite number of those variables to any desired degree of precision. So what does this mean in the continuous variable case? An arbitrary unitary transformation over even a single continuous variable requires an infinite number of parameters to define and typically cannot be approximated by a finite number of continuous quantum operations. However, it is possible to define the notion of universal quantum computation over continuous variables for various subclasses of transformations. One well considered case is the set of operations that correspond to Hamiltonians which are polynomial functions of the operators X and P of the continuous variables 29 . This set of continuous variable operations can be termed universal for the particular set of transformations if one can, through a finite number of applications of the operations, approach arbitrarily closely to any transformation in that set. However, the set cannot be too simple, or it could be efficiently classically simulated 11 . Let us start with the construction of a linear Hamiltonian of X and P of the form aX + bP + c. Basically this Hamiltonian can be constructed by applying the X operator for a short time adt, then P for a time bdt, followed by the operators P, X, —P and X for a time \/cdt. Using the relation eiAdt
iBdt
-iAdte-iBdt
_ ei[A,B}dt+0(dt3)
/^\
we can see that the net effect of these operations is our desired transformation exp [i(aX + bP + c)di\. Now by making dt sufficiently small, one can approach arbitrarily close to effecting a Hamiltonian of the desired form over small times. Then, by repeating the small-time construction t/dt times, one can approach arbitrarily close to effecting the desired Hamiltonian over time t. Thus we have a mechanism to construct any linear Hamiltonian in X and P. Next let us consider the construction of arbitrary quadratic Hamiltonians. Suppose now that one has the two quadratic Hamiltonians H = (X2 + P2) and S = (XP + PX). The Hamiltonian H is proportional to the energy operator and S is the squeezing operator. It is straightforward to show that any quadratic Hamiltonian in X and P can be constructed
209
from repeated applications of H, S, X and P using similar techniques to the linear Hamiltonian case. The action of the operators H, S, X and P in sequence only results in Hamiltonian terms that are at most quadratic. It is impossible from these operators to generate higher order terms. To construct higher order Hamiltonians, nonlinear operations such as the Kerr Hamiltonian H2 = (X2 + P2)2 are required. The Kerr Hamiltonian is a natural choice, but any nonlinear operation is sufficient. These higher order Hamiltonians have the key feature that commuting them with any Hamiltonian of the form H, S, P, X results in a Hamiltonian higher in order in X or P. For instance, the commutation between H2 and H, S, X and P allows the construction of any cubic Hamiltonian. Then the commutation between H2, H, S, X, P and these cubic Hamiltonians allows the construction of any fourth order Hamiltonian, and so on. By an iterative procedure one can construct single mode Hamiltonians that are arbitrary polynomials of X and P. However for universal quantum computation we need a multi-mode system. To construct multi-mode Hamiltonians we just require beam-splitters plus arbitrary single mode polynomials of X and P. Since the output modes of the beam-splitters are superpositions of the input modes, this allows the construction of multi-mode Hamiltonians in X and P for the respective modes. The key to this form of universal computation is the nonlinear operation and how it can be realized. Potentially this can be done using an EIT medium, or via heralded probabilistic operations and gate teleportation. In the heralded case we could use the NS gate discussed in the linear optical schemes, but with a more general signal mode and continuous variable teleportation rather than qubit teleportation. Without such nonlinear operations, our entire computation could be efficiently simulated classically. 3.1. Continuous
variables
quantum
gate
teleportation
We have previously found that qubit gate teleportation is an essential resource for single photon optical quantum computation. Continuous variable teleportation can also be used in this role, but has the significant advantage that it can be used to teleport both qubit, qudit and continuous-variable (CV) states and gates. Before we consider continuous variable gate teleportation let us review CV teleportation. Consider a three mode optical system with the modes a and b being the maximally entangled Einstein-PodolskyRosen (EPR) state |EPR) a b = / \q)a\q)bdq and mode c being an arbitrary pure state \tp). In the EPR state \q) are the position eigenstates. The tele-
210
portation protocol now works as follows: Modes a and c are subjected to the transformation joint projective measurements n , , p = Rc(q,p)\EPR)ac(EPR\Rl(q,p), where Rc(x,p)
(45)
is the Pauli operator R{q,p) = exp [-i{qp - pq)]
(46)
on mode c. Specific examples of this operator are R(q,Q) = X(q) and -R(OJP) = Z{p). Now this measurement yields two classical numbers, go and PQ. These are then used to condition the Pauli operation Rb{qo:Po), after which mode b is left in the state \tp). We can thus say that the state has been teleported from mode c to mode 6. This operation is schematically presented in Figure 11.
k>
c"vA"](qo.Po) |EPR>
/*
*
\
....':>
iR(%'P0)j— W)
Fig. 11. Schematic diagram of continuous variable quantum teleportation of state |-0) from mode c to mode 6.
We can now proceed to look at the teleportation of gate (depicted schematically in Figure 12). The process of first teleporting the state \ip) and then implementing the gate U is equivalent to acting on one mode b of the EPR state with U, followed by a modified quantum teleportation. There are a number of U gates that we can consider but as a start let U be in the single mode Clifford group. In this case we know that UR(q0,Po) 1
= R'(qo,Po)U,
(47)
where R'(qo,Po) = UR(qo,po)U~ is also an element of the Pauli group. Thus, the desired Clifford gate U can be quantum teleported onto the state |V>) simply by implementing U on one mode of the EPR pair and by appropriately altering the conditional displacement of the quantum teleportation (see Figure 12). By extension, using n single-mode quantum teleportation circuits, it is possible to teleport any gate in the n-mode Clifford group. This scheme can be used to teleport more general quantum gates than those in the Clifford group. Once we move to the more general gates we are
211
k>|EPR><*^J
ij—..
= n
|EPR>
,
RWo)-|¥'>
Fig. 12. Schematic diagram of continuous variable quantum teleportation of a gate U. The process of first teleporting the state |i/>) and then implementing the gate U, resulting in the state \xp') = U\ip) is equivalent to acting on one mode of the E P R state with U, followed by a modified quantum teleportation.
using nonlinear transformations and it is these nonlinear operations that enable universal quantum computation. For example, such a transformation could be the cubic phase gate V(j) = exp(i7<73
(48)
Let us consider the teleportation of this gate. To proceed we need to know how to commute this gate back through the Pauli operators R(qo,po)- We observe that V(-r)R(q0,PO) =
rt>(qo,Po,i)V(j),
(49)
where V^R^p^V^)-1
#2(9o,Po,7) = =
ex
P H f a o P - poq - 379o?2)] •
(50)
The term R^iqoiPOtl) is quadratic in q,p and so can be implemented with our Clifford group operations (displacements and squeezing). Thus, to teleport V(-f), this nonlinear gate is performed on the second mode of the EPR state. We follow this by a modified teleportation scheme with a conditional operation R'2(qo,Po,j) on the b mode. The result is that a state \ip) is teleported into the transformed state V(j)\ip)- Now the cubic phase gate V(-y), being a higher-order nonlinear gate on a single mode, can be combined with Clifford group gates of n modes to form a universal set of gates for QC on n modes. A scheme is known to implement this cubic phase gate, and thus with continuous variable quantum teleportation (CVQT) it is possible to teleport a universal and realizable set of gates. If the Clifford group transformations and CVQT can be implemented fault-tolerantly, then it is possible to use this scheme to implement a fault-tolerant cubic phase gate using a gate that is not fault-tolerant. This is the key result and shows that any nonlinear transformations can be moved "off-line". Although these transformations must still be performed in the quantum teleportation circuit,
212
they can be made to act on EPR ancilla states non-deterministically rather than on the fragile encoded states. Lastly because most optical quantum information schemes employ the Kerr effect (generated by a Hamiltonian of the form {a))2a?) as the nonlinear transformation outside of the Clifford group, it is of interest to consider how such a transformation can be implemented using the above univer2 sal set of gates. Using the relation e^teiBt£-iAte-iBt = ei[A,B}t + Q{f^ a combination of cubic phase gates and Clifford gates can be used to simulate the Kerr nonlinearity to any degree of accuracy. 4. Hybrid schemes: combining single photon logic and continuous variables Now that we have considered quantum computation using single photons and continuous variables, let us examine hybrid schemes where we use elements from both. Our primary motivation here is to encode the quantum information using the single photons and to use the continuous variables as the communication channel between the qubits. The key elements in this hybrid scheme are weak cross-Kerr nonhnearities, too weak to realise a CPhase gate directly. 4.1. A QND
detector
Before we begin our discussion of the construction of efficient quantum gates using weak nonhnearities, let us first consider a photon number quantum non-demolition (QND) measurement using a cross-Kerr nonlinearity 30 . The cross-Kerr nonlinearity has a Hamiltonian of the form HQND = hxhahc
(51)
with the signal (probe) mode having the photon number operators ha = a^a {hc = c*c) where a\a {c*,c) are the respective creation and annihilation operators for the modes and \ ls the strength of the nonlinearity. This Hamiltonian is photon-number preserving, that is, it does not change the photon number in either the signal or probe modes. This Hamiltonian enables a unitary transformation of the form UQND = exp [-i6hahc]
(52)
to be implemented where 9 = \t with t being the interaction time between the signal and probe beams with the Kerr material. Let us now consider its action on the signal and probe modes. If the signal field initially contains na photons and the probe field is in an initial
213
coherent state with amplitude ac, the cross-Kerr nonlinearity causes the combined system to evolve as |*(*)}out = UQND\na)\ac)
=
\na)\acem«a).
(53)
The Fock state \na) is unaffected by the interaction with the cross-Kerr nonlinearity but the coherent state \ac) picks up a phase that is directly proportional to the number of photons na in the signal state \na). If we measure this phase shift we can then indirectly infer the number of photons in the signal mode a. This can be achieved simply with a homodyne measurement (depicted schematically in Figure 13).
|ac>
in0\ acemH >
!
Cross Kerr Nonlinearity
wm^^
e
I
Homodyne Measurement
o>
Signal In
Signal Out
Fig. 13. Schematic diagram of a photon resolving detector based on a cross-Kerr nonlinearity and a homodyne measurement. The two inputs are a Fock state \na) (with n a = 0,1,..) in the signal mode a and a coherent state with real amplitude ac in the probe mode c. The presence of photons in mode a causes a phase shift on the coherent state \ac) directly proportional to n a , which can be determined with a momentum quadrature measurement on mode c.
The homodyne apparatus allows measurement of the quadrature operator x(4>) = ce l * + c^e _l * with <j> being the phase of the local oscillator. This has the expectation value (x((j))) = 2Re [ac] cosS
(54)
where 5 =
214
quadratures yield the expectation values (X) = 2ac cos {na6) (Y)=2ac
sin (n„0)
(55)
with a unit variance. If the inputs in mode a are the Fock state |0) or |1), the respective outputs of the probe mode c are either the coherent states \ac) or \ace%e). The probability of misidentifying these states is given by P
1
= -Frfc
error — ^ •
LJi
J-^
SNRy 2\/2 .
(56)
where SNRy = 2acsm(na9). A signal to noise ratio of SNRy = 2n would thus give .Perror ryJ -10 ^ and hence we require otcsin0 ^ 7r. This can be achieved with a small nonlinearity 9 as long as the probe beam is intense enough 3 0 . 4.2. A polarization-preserving
QND
detector
In many optical quantum computation tasks our information is not encoded in photon number but in polarization instead. When our information encoding is polarization-based there are two separate detection tasks that we need to perform. The first and simplest is just to determine, for instance, the polarization (\H) or \V)) of a photon. This can be achieved by converting the polarization information to "which-path" information on a polarizing beam-splitter. The "which-path" information is photon-number encoded in each path and hence a separate QND photon number measurement of each path will determine which polarization basis state the photon was originally in, or project it into this basis. This can be thought of as two applications of the QND detector. The second task (one that is critically important for error correction codes in optics) is to determine whether our single-photon polarizationencoded qubit is present or not. That is, for the optical field under consideration we want to determine whether it contains a photon or not. If it does contain a photon, we do not want to destroy the information in its polarization state. This can be achieved by first converting the polarization qubit to a "which-path" qubit. Each path then interacts with a weak crossKerr nonlinearity 9 using the same shared probe beam (Figure 14). If a photon is present in either path of this signal beam it induces a phase shift 9 on the probe beam; however, with this configuration it is not possible to determine which path induced the phase shift.
215
Fig. 14. Schematic diagram of a polarization-preserving photon number quantum nondemolition detector based on a pair of identical cross-Kerr optical nonlinearities. T h e signal mode is a Fock state with an unknown polarization, which is converted into whichpath encoding by a polarizing beam splitter (square box). T h e phase shift suffered by the probe mode is proportional to na, independent of the polarization of the signal mode photon.
To illustrate this, consider a signal mode initially prepared in the state co|0) + CH \H) + cy|V). The action of the polarizing beam-splitter converts this polarization-encoded state to the which-path-encoded state co|00) + c#|10) + cy |01). Now the interaction between the path mode, probe field and the first nonlinearity creates the three mode state co|00)|a) p + cH| 10}\ae i8 ) p + cy|01)|a) p .
(57)
The second weak nonlinearity causes a phase shift on probe beam component of the last term. This results in the three-mode system evolving to co|00)|a) p + [cH|1Q) + cv |01>] \aeie)p .
(58)
The last polarizing beam-splitter changes the which-path encoding back to the polarization encoding, resulting in co|0)|a) p + [cH\H) + cv\V)} \aei0)p .
(59)
It is now very clear that if a phase shift is detected, we know that a single photon is present in the signal mode without destroying its polarization encoding. The real key in this setup is the use of separate but identical cross-Kerr nonlinearities which allows different paths to interact with the same probe beam, each potentially inducing a phase shift on it. If a phase shift is detected we can infer the presence of a photon, but the coherence between the paths is perserved. Finally it is enlightening to ask what happens if phase shifts from the different paths are not identical. If we consider the above example with the upper path inducing a phase shift 9 and the lower path inducing —9 phase shift, then the resulting three-mode state is co|0)|a) p + cH\H)\aeie)p
+ cv \V)\ae-id)p.
(60)
216
This is interesting because the homodyne measurement now allows us to distinguish all three states |0), \H) and |V) without having to use two full QND detectors. Only one probe beam and homodyne measurement is required. 4.3. A two-qubit
parity
gate
The concept of using multiple cross-Kerr nonlinearities is very interesting, as we have seen in the previous section. However, there is no reason to require the probe beam to interact only with one photonic qubit. This detection concept could be applied to multiple qubits. If we want to perform a more "generalized" type of measurement between different photonic qubits, we could delay the homodyne measurement, having the probe beam interact with several cross-Kerr nonlinearities where the signal mode is different in each case, as in Figure 15. The different signal modes could be from separate photonic qubits. The probe beam measurement then occurs after all these interactions, in a collective way which could, for instance, allow a non-destructive detection that distinguishes superpositions and mixtures of the states \HH) and \VV) from \HV) and \VH). The key here is that we could have no net phase shift on the \HH) and \VV) terms whilst having a phase shift on the \HV) and \VH) terms. We will call this generalization a two-qubit polarization parity QND gate. Let us consider two polarization-encoded qubits given by \*12) = (30\HH) + falHV) + (32\VH) + (33\VV) .
(61)
We have written this as the most general two qubit pure state. Now consider the action of the first weak nonlinearity on the \H) component of the first qubit. The state of the system (both qubits and probe beam) is p0\HH)\aceid)p
+ (31\HV)\acei9)p
+ p2\VH)\ac)p
+ fo\VV)\ac)p . (62)
Next, application of the second cross-Kerr nonlinearity on the \H) component of the second qubit leads to |V)r = \Po\HH) +P3\VV)}\ac)p
+ 01\HV)\aceie)p
+ (32\VH)\ace-ie)p
.
This state is illustrated in the phase-space plot in Figure 16. It is now obvious that the \HH) and \VV) terms pick up no phase shift and remain coherent with respect to each other while the \HV) and \VH) pick up opposite sign phase shifts of 9. We thus need to perform a measurement that does not allow the sign of the phase shift to be determined. An X homodyne measurement achieves
217 Homodyne Measurement
Probe
classical feedforward
4
|HV)
1
X quadrature 7*|HH)+|W) / |VH)
Fig. 16. Schematic phase-space illustration of the three mode quantum state \ip)r = \J3o\HH) +fo\VV)] \ac)p+Pi\HV)\aceie)v+p2\VH)\ace-i6)p. For the \HH) and \VV) components of the state, the probe beam does not suffer from a phase shift, while for the \HV) and \VH) components, the probe beam acquires phase shifts of 6 and - 0 respectively.
this 31 by projecting the probe beam to the position quadrature eigenstate |X)(X| with a measurement result X. The resulting two qubit state is
218
then 31 \iPx) = l0o\HH)+p3\VV)}
(X\ac)p
+
fo\HV){X\acei6)p
+
i6
(63)
P2\VH)(X\ace- )p
Now using the expression (x\a)
(27T) 1 / 4
exp
-Im(a) 2 -
[x - 2a) 2
(64)
we obtain \ipx) = f(X,ac)\fo\HH) + f(X,ac
+ fo\VV)}
(65)
+ foei
cos 6) PieWx^HV)
where, assuming that ac is real f(x,(3)
= e x p -\(x-2P)2
/(?< ,1/4
(66)
(67)
Now f(X, ac) and f(X, ac cos 6) are two Gaussian curves with the mid point between the peaks located at X0 = ac [1 + cos 6] and the peaks separated by a distance Xd = 2a c [1 — cos#]. As long as this difference is large Xd ~ ac92 3> 1, then there is little overlap between these distributions and so our measurement can project the system into one of the two subspaces. The probability of a discrimination error occurring is given by = iErfc[X d /2v / 2]
(68)
which is less than 10 4 when Xd > 2TT. For a measurement result X such that X > Xo, our solution (65) collapses to the even parity state \^x>x0)^fo\HH)+fo\W)
(69)
while for X < Xo we get the odd parity state |^x<x 0 ) ~ foe*™\HV) +
foe-«W\VH)
.
(70)
The action of this two mode polarization non-demolition parity gate is clear. It splits the even parity terms (69) nearly deterministically from the odd parity cases (70). The odd parity state (70) depends on the value of the measured quadrature X. Simple local rotations using phase shifters that dependent on the measurement result X can be performed via a classical feedforward process to transform to \1>x<xo)~0i\HV)
+ fo\VH)
(71)
219
which is independent of X. This correction operation requires that we are able to precisely determine X and so accurately determine the phase correction
Fig. 17. Schematic diagram of a two-qubit polarization parity gate which does require the measurement-dependent phase correction.
Consider a two-qubit polarization-encoded state (61). After the interaction with the first two weak nonlinearities (plus a —9 phase rotation on the probe beam), our combined two qubit and probe system has the form \i/>) = \fa\HH) +fo\VV)]
K ) p + /?1|W)|acei<% +
02\VH)\ace^eU7.2)
We now displace the probe beam by an amount — 2ac cos 6 and the system evolves to \1>) = IP0\HH)+{33\VV)}
| o c ( l - 2cos#)) p
+ 0ieiciz sin 20j HV)\ac(eie + 026
-ial
sm2d
(73)
-2cosB))p i6
VH)\ac(e-
-2cos9))p.
Now applying the last two weak cross-Kerr nonlinearities and a phase shift —6 on the probe beam gives \i>) = \po\HH) + (33\VV)} | a c ( l - 2cos0)) p
+
Pxeia2-cos2e\HV)
+ (i2e-ia"cos28\VH)
| - ac)p
(74)
We see immediately that our probe beam now has only tv/o state components, | — ac)p and |a c (2 cos 9 — l)) p , and hence an appropriate probe
220
measurement will project us into the even or odd parity qubit subspace, without the need for a measurement-dependent phase correction. The cost is that two extra weak nonlinearities are needed. We do have a phase shift on the odd parity subspace that needs to be removed, but this can be done easily as it does not depend on the measurement outcome. It is also useful to note that the separation between the two coherent states scales as ac92 for 8
+p2e-ia*cos2e\VH)]
\ac)p
(75)
before the measurement. Repeating the circuit in Figure 17 with the additional final displacement 2ac gives |V) = \J3o\HH) + fo\VV)] \ac (1 + 4(1 - cos0))) p + ^ieia'cos2e\HV)
+p2e-ia*cos26\VH)'j
\ac)p .
(76)
Repeating this circuit a total of n times gives |V) = \po\HH) + f33\VV)} \ac (1 + 2n(l - cosfl))),
+
/3iema<:cos2e\HV)+f32e-ma°cos2y\VH)
\ac)p
(77)
and so we see our two coherent states are now separated by 2nac (1 — cos#) ~ nac62. We see that the use of the circuit n times allows us to increase the separation by a factor of n. The total number of nonlinearities used is now An but this has the advantage that ac can be smaller. So far all of our measurements on the probe beam have used standard homodyne measurements. Whilst these measurements are in principle straightforward to implement, they are by no means optimal. A nearoptimal measurement is preferable as this enables the strength of the crossKerr nonlinearities used to be near-minimal 44 . Let us consider the combined state of the system in Figure 15 before the measurement and apply a displacement operation D(—a) on the probe beam. This results in the state \V)abp = [Po\HH)ab + /33\VV)ab] |0) p + e - ^ ^ A I t f ^ U M e ^ - 1))„ +
ia2 ine
ie
e * p2\VH)ab\ac(e- -l))p.
(78)
221
The \HV)ab and \VH)ab amplitudes have picked up a phase shift due to the displacement, which can be simply removed using linear optical elements. As long as ac6 > IT, a QND photon number measurement on the probe beam will project the a and b mode state into mab
= p0\HH)ab
+ (33\VV)ab
(79)
for a QND photon number measurement np of zero and |*>ab = /3iei*
(80)
for rip > 0 where <^(np) ~ — np^. The error in discriminating the two components (even and odd parity states) is approximately Perr ~ 1 0 - 4 for a9 ~ 7r. This is a near optimal measurement and can be realised if one has a weak nonlinearity and efficient homodyne measurements 44 . For all three approaches above we have chosen to call the even parity state {\HH),\VV)} and the odd parity states {\HV),\VH)}, but this is an arbitrary choice, primarily dependent on the form or type of PBS used to convert the polarization encoded qubits to which path encoded qubits. Any other choice is also acceptable and it does not have to be the same between the two qubits. 5. Bell state measurements These near-deterministic non-destructive parity measurements are critically important in optical quantum information processing. They can be used to build a number of interesting devices and gates. For instance, with a nondestructive parity gate it is possible to build a Bell-state detector. Bell-state measurements are known to be one of the essential and enabling tools in quantum computation and communication. If we consider the four Bell states \*+) = -L(\H,V)
+ \V,H))
(81)
\9-) = l=(\H,V)-\V,H))
(82)
\*+) = ±(\H,H)
(83)
+ \V,V))
\$-) = ±(\H,H)-\V,V))
(84)
then it is clear that our standard parity gate will distinguish | $ ± ) from | * ± ) . Hence an application of the parity detector distinguishes two subclasses of the Bell states.
222
The natural question now becomes: can we modify the parity gate to distinguish | * + ) from I * - ) and | # + ) from | $ - ) ? In the parity gate used above the {H, V} polarizing beam-splitters dictates that {\H,H),\V,V}} are distinguished from {\H,V),\V,H)}. Bell states can be written in the {D = H + V,D = H-V} basis as \H,V) + \V,H)^\D,D)-\D,D)
(85)
\H,V)-\V,H)-*\D,D)-\D,D)
(86)
\H,H) + \V,V)^\D,D) + \D,D)
(87)
\H,H)-\V,V)-*\D,D) + \D,D)
(88)
+
With respect to {D, £>}, |*+) and | $ ) are even parity and \$>~) and |$~) are odd parity. Hence the {D, D} parity gate will allow us to distinguish
{|*+), |* + )} from (I* - ), I* - )}.
Fig. 18. Schematic diagram of a non-destructive Beli state measurement, composed of two QND parity detectors. The first parity gate uses the standard {H, V} PBS and distinguishes the I**) Bell states from the I**) ones. An even-parity result for gate indicates the presence of 1$*} while an odd-parity result indicates the presence of I**}. For the latter result a local operation on the second qubit is required to remove the phase $(Xi) induced by the measurement. Once this correction is done the second parity gate can be applied. This gate is similar to the first one but has 45-deg PBS's (square box with circle inside) that operate in the {D, D} basis. An even-parity result indicates the presence of | # + ) or \'$+) while an odd parity result indicates the presence of the |
Now since the both ({H, V} and {D, D}) parity gates are nondestructive on the qubits and select different pairs of Bell states, they allow a natural construction of a Bell-state analyser (depicted in Figure 18). From each parity measurement we get one bit of information, indicating whether the parity was even or odd, and so from both parity measurements we end up with four possible results (even, even), (even, odd), (odd, even) and (odd, odd). This is enough to uniquely identify all the Bell states. It is important to remove the unwanted phase factors that have arisen after an odd parity measurement result. This needs to be done in the same basis as the PBS
223
in the particular parity gate. For instance, for an odd parity result giving X = X\ on the first parity gate, a phase shift (j>{X{) needs to be removed in the {H, V} basis. Similarly for a odd parity results giving X = X2 on the second parity gate a phase shift (f>(X2) needs to be removed in the {D, D} basis. So far we have shown how it is possible, using linear elements, weak cross-Kerr nonlinearities and homodyne measurements, to create a wide range of high efficiency quantum detectors and gates that can perform tasks ranging from photon-number discrimination to Bell-state measurements. This is all achieved non-destructively on the photonic qubits and so provides a critical set of tools, extremely useful for single-photon quantum computation and communication. With these tools, universal quantum computation can be achieved using the ideas and techniques originally proposed by KLM. 6. A resource-efficient C N O T gate The parity gate and the Bell-state analyser have shown the versatility of an approach using weak nonlinearities and homodyne conditioning measurements. Both of these gates/detectors can be used to induce two-qubit operations and hence are all that is necessary, with single-qubit operation and single-photon measurements, to perform universal quantum computation. However, the parity and Bell-state gates are not the typical two-qubit gates that one generally considers in the standard quantum computational models. The typical two-qubit gate generally considered is the CNOT gate. This can be constructed from two parity gates (like the Bell-state detector) but requires an ancilla qubit. This CNOT gate is depicted schematically in Figure 19) and operates as follows: Suppose that the control and target qubits are initially in the joint state c0\HH)ct
+ a\HV)ct
+ c2\VH)ct
+ c3\VV)ct
(89)
with the ancilla mode prepared in |i/)o + |V) a . The action of the first parity gate between the control and ancilla mode (c and a) conditions the system to c0\HHH)cat
+ Cl\HHV)cat
+ c2\VVH)cat
+ c3\VVV)cat
(90)
for an even-parity result and coe^x^\HVH)cat
+ c1ei^x^\HVV)cat
+
ae-^^^VHH)^ x
+ c3e-^ ^\VHV)cat
(91)
224
!<*<:> Control Out Control in Ancilla in
*c+o*
Target in |Oc>
N 4*2 i
^^i^
MH '•.3'" Target Out
Fig. 19. Schematic diagram of a near-deterministic CNOT composed of two parity gates (one with PBS in the {H, V} basis and one with PBS in the {D, D} basis), one ancilla qubit prepared initially in \H) + \V), a polarization-determining photon-number QND measurement and classical feedforward elements.
for an odd-parity result. The unwanted phase
Cl)
\H)C + (c 2 - c3) \V)e] \DD)at
(92)
+ [(co + c,) \H)C - (ca + c3) \V)C] \DD)at . The action of the {D, D} parity gate is now clear. For an even-parity result the above state is projected to [(co +
Cl)
\H)C + (ca + c3) |V)C] \DD)at
+ [(co - cx) \H)C - (c2 - c3) \V)C] \DD)at
(93)
while for an odd-parity result (after phase correction) we obtain [(co -
Cl)\H)c
+ (c2 -
c3)\V)c]\DD)at
+ [(co + ci) \H)C - (c2 + c3) \V)C] \DD)at .
(94)
Now this odd parity state can be transformed to the even parity case by bitflipping the ancilla qubit and performing a sign flip on the IV') component of the control. After such an operation, the state is given by (93). Now performing a measurement of the ancilla mode in the {H, V} basis, the control and target qubits are transformed to co\HH)ct + Cl\HV)ct
+ c2\VV)ct
+ c3\VH)ct
(95)
225
for an H result, and co\HV)ct + a\HH)ct
+ c2\VH)ct
+ c3\VV)ct
(96)
for a V result. The second case can be transformed to the first by a bit flip of the target qubit. After all such operations and phase corrections, our initial control and target qubits have been transformed as c0\HH)ct
+ cx\HV)ct
+ c2\VH)ct
+ c3\VV)ct
- co\HH)ct + Cl\HV)ct
(97) + c2\VV)ct
+ c3\VH)ct
.
This is clearly the result one would expect if a CNOT operation had been performed on the control and target qubits and shows how our weak nonlinearity approach can implement a near deterministic CNOT operation, utilizing only one ancilla qubit (which is not destroyed at the end of the gate and could in principle be re-used). This represents a huge saving in the physical resources to implement single-photon quantum logic gates, compared to the previous linear optics schemes. 6.1. A discussion
on the weak nonlinearity
approach
We have shown how it is possible to create near deterministic two-qubit gates (parity, Bell and CNOT) without a huge overhead in ancilla resources. In fact, an ancilla photon is required only for the CNOT gate. The key additions to the general linear optical resources are weak cross-Kerr nonlinearities and efficient homodyne measurements. Homodyne measurements are a well established technique, frequently used in the continuous variable quantum information processing community. However weak cross-Kerr nonlinearities are not commonly used within optical quantum computational devices and as such a discussion of the source and strength of such elements is required. Before this we really need to define what we mean by weak— weak compared with what? Basically it is well known that deterministic two-qubit gates can be performed if one has access to a cross-Kerr nonlinearity that can induce a 7r phase shift directly between single photons. This leads to a natural definition of weak nonlinearities, that is, the use of nonlinear cross-Kerr materials (when all are taken into account) that cannot directly induce a phase shift of the order of ir. This seems to give an acceptable functional definition. For the majority of the parity gates discussed previously we have established that the nonlinearity 6 must satisfy the constraint ac62 ~ 8 where ac is the amplitude of the probe beam. For a weak nonlinearity ^ < 1 we must choose ac ~ 10/8 2 , so for instance if 9 ~ 10~ 2 then ac > 105
226
(which corresponds to a probe beam with mean photon number 10 10 ). For a smaller 6 we need a much larger ac. This puts a natural constraint on 0, since ac cannot be made arbitrarily large in practice. For the most efficient parity-based gates discussed previously we have established that the nonlinearity 9 must satisfy the constraint a6 ~ IT where, just to re-emphasize, a is the amplitude of the probe beam. Thus, due to the weak nature of the nonlinearity S < 1 we must choose a ~ TT/9, so for instance if 9 ~ 10~ 5 then a > 7rl05 (which corresponds to a probe beam with mean photon number 10 11 ). This leads to the question of a mechanism to achieve the weak cross-Kerr nonlinearity. Natural x3 materials have small nonlinearities 32 on the order of 1 0 - 1 8 which would require lasers with ac ~ 10 37 , which is physically unrealistic. However, systems such as optical fibers33, silica whisperinggallery micro-resonators 34 , cavity QED systems 35,36 and EIT 3 7 are capable of producing much larger nonlinearities. For example, calculations for EIT systems in NV diamond 30 have shown that potential phase shifts of order 9 = 0.01 are achievable. With 9 = 0.01 the probe beam must have an amplitude of at least 103, which is physically reasonable with current technology. Finally, by using these weak cross-Kerr nonlinearities to aid the construction of near deterministic two-qubit gates, we can build quantum circuits with far fewer resources than are required for the current corresponding approaches with linear optics. This has enormous implications for the development of single-photon quantum computing and information processing, using either the conventional gate-based models, or cluster-state techniques. The approach can be applied directly to optical cluster-state computation, allowing a significant reduction in the physical resources needed. It is straightforward to show that in principle an n-qubit computation only requires of order n single photon sources. This truly indicates the power of a weak nonlinearity approach. The strength of the nonlinearities required for our gate are also orders of magnitude weaker than those required to perform CNOT gates directly between the single photons. 7. Concluding discussion: is optical computation possible? We hope that so far we have shown the progress in using optical systems for quantum computation (QC) and quantum information progressing (QIP). Optical QIP is currently a very active research area, both theoretically and experimentally. To be able to perform universal quantum computation with optical fields (either at the single photon level or continuous variable level)
227
it is known that nonlinearities are needed. The form of the required optical nonlinearities doesn't really matter; for example they can be generated via measurement. The work of Knill, Laflamme and Milburn (KLM) has shown that in principle universal quantum computation is possible with linear optics 17 and using such nonlinearities. However, due to the probabilistic nature of the gates in linear optical QIP, it is practically rather inefficient (in terms of photon resources) to implement 39-41 . Strong cross-Kerr nonlinearities are able to effectively mediate an interaction directly between photonic qubits. This would realize deterministic quantum gates and thus efficient optical QIP. However, in practice, such nonlinearities are not available. On the other hand, much weaker nonlinearities can be generated, for example, with electromagnetically induced transparency (EIT) 3 7 ' 3 8 ' 3 0 . We have shown how such weak nonlinearities provide the building blocks for efficient optical Qlp35,3i,42-46 There has also been a significant amount of experimental progress in demonstrating optical gates for quantum computation. The last five years has seen a number of key demonstrations, including: • • • • • • • • • •
A destructive two-photon parity and CNOT gate 47 ' 48 A three-photon destructive CNOT gate 49 The nonlinear sign-shift gate 50 The Franson four-photon CNOT gate 51,52 A complete Bell-state measurement 53 Teleportation and entanglement swapping with polarization-encoded qubits 54 ' 55 Generation of the three- and four-photon cluster states 56,57 Encoding qubits for error prevention 58 ' 59 Continuous-variable teleportation and entanglement swapping 60-63 The generation of weak nonlinearities and the slowing of light 36 ' 64
All of these demonstrations show the potential of optics for information processing. What we have not discussed so far is the generation of relevant initial quantum states and their detection. To do this we need to consider the single-photon and continuous-variable situations separately. We have discussed the single-photon detection requirement previously, but not how this can be achieved. In the single-photon qubit regime we need two critical resources to enable scalable quantum computation: a generator of single photons on demand and a detector to measure whether a photon is present or not. With these resources and linear optical elements, it is straightforward to create
228
an arbitrary polarization-encoded qubit and to detect it. The key is the generation of the on-demand source and the photon detection. Currently both of these resources are still in development65 but are not efficient enough yet. However, the progress is very promising. In the continuous variable regime the generation of an initial state and the measurement of the qunat in the computational basis are rather straightforward. If we encode our qunat in terms of the X and P quadratures of the electromagnetic field, then for the initialization we need to be able to generate a position or momentum quadrature eigenstate. The vacuum state |0) is ideal for this, and simply generated. The measurement of the qunat in the computational basis can be achieved via a highly efficient homodyne measurement (depicted schematically in Figure 20). The homodyne measurement works as follows: Consider a signal mode specified by the creation and destruction operators a* and a and an intense local oscillator celB where 6 is the local oscillator phase.. If the signal mode and oscillator are mixed on a 50/50 beam-splitter and then the output mode intensities are measured on photodiodes, it is straightforward to show that the difference in the currents from the detectors C and D is Id = (etc) - (dU) = r,e(a^e~ie + aeie) = rjeXe
(98)
where Xo = X and Xv/2 = P, and rj is the efficiency of the detectors. Thus as long as one knows rj and e, a measurement of the current difference measures X or P depending on the local oscillator phase. The homodyne measurements currently have overall efficiencies exceeding 99% 66 . Now to complete our discussion of optics for quantum computation let us summarize how well the DiVincenzo criteria are satisfied, for both our single-photon and continuous-variable variable computation schemes. 7.1. The DiVincenzo
criteria
for single
photons
For single-photon quantum computation the status of the DiVincenzo criteria is as follows: • We have well defined qubits in terms of the polarization degree of freedom of single photons, which is equivalent to which-path encoding. • The initialization of qubits corresponds to creating single-photon states with well-defined polarizations. Such single-photon sources are cur-
229
1 Signal
Homodyne /^Detector
7 ^-D
t
Local Oscillator Fig. 20. Schematic diagram of a homodyne measurement. The homodyne detector is composed of two optical modes: a signal mode and an intense local oscillator. Both these modes are combined on a 50/50 beam-splitter and the difference in photocurrents between the detectors on the output modes is measured.
rently under active development and are promising: At this moment, there exist sources that have sufficient coherence for self-interference. • Photons have extremely long coherence times, and suffer from low loss when traveling through optical fibers or free space. • For universal computation we need photons to interact with each other. A two-qubit interaction is essential. In free space photons do not interact with each other; however, we can induce a probabilistic two-photon gate by using the bunching effect at beam-splitters and projective measurements, or via nonlinear effects (such as those generated by EIT). These nonlinear effects do not necessarily need to be strong; weak nonlinearities with a coherent field acting as a bus can be used to implement near deterministic CNOT gates between photonic qubits. Alternatively the one-way or cluster-state quantum computational models can be used. • Next, we need the ability to faithfully read the quantum state of a qubit in the computational basis. In the single-photon case this is achieved through photodetection. However, current detectors have less than ideal properties. Typical quantum efficiencies reach at most 90 percent, while we need at least 99 percent. There are, however, exciting new proposals for high-fidelity and high efficiency photodetectors, which don't even absorb the photons.
230
The two additional criteria of DiVincenzo are easily satisfied in singlephoton optical quantum computing. Since single photons can be considered as "flying qubits", there is no difference between the flying qubits needed for quantum communication and the "stationary qubits" used for computation. Photons can be transmitted between remote locations. Static memory for photonic quantum information may be a useful asset in the future, for large scale quantum processing. 7.2. The DiVincenzo
criteria
for continuous
variables
For continuous variables we may summarize the status of the DiVincenzo criteria as follows: • We need to relax the criterion for qubits. We require a collection of well-characterized qu-units (qu-bit, qu-dit, and qunat). For continuousvariables (qunats) we can use X and P (the quadratures) of the electromagnetic fields to encode our quantum information. • State preparation is straightforward as the vacuum of the electromagnetic field can be used. No preparation is required to create the vacuum. • Optical systems tend to have long coherence times. However the loss of a single photon in CV systems can be problematic. The exact coherence properties depend on the physical implementation, that is whether it is in free space or in a fiber. • We need to define what we mean by quantum computation, as a continuous variable needs an infinite number of parameters to describe it. We demand that we can create Hamiltonians that are polynomial functions of the operators corresponding to the continuous variables. Such Hamiltonians should be of degree higher than quadratic (if homodyne measurements are used), otherwise the system could be efficiently classically simulated. • The choice of measurement is homodyne/heterodyne measurements. These are projective measurements in the X (and/or P) basis and have been implemented with very high efficiency. For optical frequencies the efficiencies exceed 99 percent at present 66 . So to finally conclude, we believe that optical systems are ideally suited for quantum communication and are a very promising candidate for quan-
231
turn computation. It is very likely t h a t fully q u a n t u m information processing devices will contain optical components at their core. 8.
Acknowledgments
We t h a n k S. D. Barrett, R. G. Beausoleil, P. Kok and G. J. Milburn for numerous valuable discussions. This work was supported in p a r t by Japanese J S P S , M P H P T , and Asahi-Glass research grants and the E u r o p e a n Project RAMBOQ. References 1. M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information, (Cambridge University Press, Cambridge, U.K., 2000); H.-K. Lo, S. Popescu and T. P. Spiller (eds.), Introduction to Quantum Computation and Information, (World Scientific Publishing, 1998). 2. N. Gisin, G. Ribordy, W. Tittel, and H. Zbinden, Rev. Mod. Phys. 74, 145 (2002). 3. R. G. Clark (ed.), Experimental Implementation of Quantum Computation, (Rinton Press, 2001), ISBN 1-58949-013-4. 4. N. A. Gershenfeld and I. L. Chuang, Science 275, 350 (1997); D. G. Cory, A. F. Fahmy, and T. F. Havel, Proc. Nat. Acad. Sci. 94, 1634 (1997); Y. Nakamura, Yu. A. Pashkin and J. S. Tsai, Nature 398, 786 (1999); D. Vion, A. Aassime, A. Cottet, P. Joyez, H. Pothier, C. Urbina, D. Esteve, and M. H. Devoret, Science 296, 886 (2002); Y. Yu, S. Han, X. Chu, S. Chu and Z. Wang, Science 296, 889 (2002); B. E. Kane, Nature 393, 133 (1998). 5. See for instance: P. G. Kwiat, K. Mattle, H. Weinfurter, A. Zeilinger, A. V. Sergienko and Y. Shih, Phys. Rev. Lett. 75, 4337 (1995); W. Tittel, J. Brendel, H. Zbinden and N. Gisin, Phys. Rev. Lett. 8 1 , 3563 (1998); P. G. Kwiat, E. Waks, A. G. White, I. Appelbaum and P. H. Eberhard, Phys. Rev. A 60, R773 (1999); G. Weihs, T. Jennewein, C. Simon, H. Weinfurter and A. Zeilinger, Phys. Rev. Lett. 8 1 , 5039 (1998). 6. D. P. DiVincenzo, "Topics in Quantum Computers", in Mesoscopic Electron Transport, (ed. Kowenhoven, L., Schn, G. & Sohn, L.), NATO ASI Series E, (Kluwer Ac. PubL, Dordrecht, 1997); cond-mat/9612126. 7. D. P. DiVincenzo, Fortschr. Phys. 48, 9 (2000). 8. P. W. Shor, Phys. Rev. A 52, 2493 (1995); P. W. Shor, "Fault-tolerant quantum computation", Proc. 37th Annual Symposium on the Foundations of Computer Science, 56 (IEEE Computer Society Press, Los Alamitos, CA, 1996), quant-ph/9605011; A. M. Steane, Phys. Rev. Lett. 77, 793 (1996); A. M. Steane, Phys. Rev. Lett. 78, 2252 (1997). 9. D. Gottesman, in Proceedings of the XXII International Colloquium on Group Theoretical Methods in Physics, edited by S. P. Corney et al. (International Press, Cambridge, MA, 1999), p. 32. 10. M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information, (Cambridge University Press, Cambridge, U.K., 2000), p 464.
232
11. S. D. Bartlett, B, C. Sanders, S. L. Braunstein and Kae Nemoto, Phys. Rev. Lett. 88, 097904 (2002). 12. T. Rudolph and L. Grover, A 2 rebit gate universal for quantum computing, quant-ph/0210187. 13. M. A. Nielsen, Phys. Lett. A 308, 96 (2003). 14. D. W. Leung, Int. J. Quant. Inf. 2, 33 (2004). 15. R. Raussendorf and H. J. Briegel, Phys. Rev. Lett. 86, 5188 (2001). 16. Kae Nemoto and W. J. Munro, Universal quantum computation on the power of quantum non-demolition measurements, accepted to Phys. Lett A (2005). 17. E. Knill, R. Laflamme and G. Milburn, Nature 409, 46 (2001). 18. T. C. Ralph, A. G. White, W. J. Munro and G. J. Milburn, Phys. Rev. A 65, 012314 (2002). 19. T. Rudolph and Jian-Wei Pan, A simple gate for linear optics quantum computing, quant-ph/0108056. 20. E. Knill, Phys. Rev. A 66, 052306 (2002). 21. T.B. Pittman, B.C. Jacobs and J.D. Franson, Phys. Rev. A 64, 062311 (2001). 22. D. Gottesman and I. L. Chuang, Nature 402, 390 (1999). 23. C. H. Bennett, G. Brassard, C. Crepeau, R. Jozsa, A. Peres, and W. K. Wootters , Phys. Rev. Lett. 70, 1895 (1993). 24. M. Nielsen, Phys. Rev. Lett. 93, 040503 (2004). 25. D. E. Browne and T. Rudolph, Phys. Rev. Lett. 95, 010501 (2005). 26. R. G. Beausoleil, W. J. Munro and T. P. Spiller, J. Mod. Opt. 5 1 , 1559 (2004). 27. S. E. Harris and Y. Yamamoto, Phys. Rev. Lett. 8 1 , 3611 (1998). 28. A. Gilchrist, G. J. Milburn, W. J. Munro and Kae Nemoto, Generating optical nonlinearity using trapped atoms, quant-ph/0305167. 29. S. Lloyd and S. L. Braunstein, Phys. Rev. Lett. 82, 1784 (1999). 30. W. J. Munro, K. Nemoto, R. G. Beausoleil and T. P. Spiller, Phys. Rev. A 71, 033819 (2005). 31. S. D. Barrett, P. Kok, Kae Nemoto, R. G. Beausoleil, W. J. Munro and T. P. Spiller, Phys. Rev. A 7 1 , 060302 (2005). 32. P. Kok, H Lee and J. P. Dowling, Phys. Rev. A 66, 063814 (2002). 33. X. Li, P. L. Voss, J. E. Sharping and P. Kumar, Raman-noise induced quantum limits for \ nondegenerate phase-sensitive amplification and quadrature squeezing, quant-ph/0402191. 34. T. J. Kippenberg, S. M. Spillane and K. J. Vahala, Phys. Rev. Lett. 93, 083904 (2004). 35. P. Grangier, J. A. Levenson and J.-P. Poizat, Nature 396, 537 (1998). 36. Q. A. Turchette, C. J. Hood, W. Lange, H. Mabuchi and H. J. Kimble, Phys. Rev. Lett. 75, 4710 (1995). 37. H. Schmidt and A. Imamoglu, Optics Letters 2 1 , 1936 (1996). 38. M. Paternostro, M. S. Kim and B. S. Ham, Phys. Rev. A 67, 023811 (2002)). 39. S. Scheel, Kae Nemoto, W. J. Munro and P. L. Knight, Phys. Rev. A 68, 032310 (2003). 40. S. Scheel, Scaling of success probabilities for linear optics gates, quant-
233
ph/0410014. 41. J. Eisert, Optimizing linear optics quantum gates, quant-ph/0409156. 42. Kae Nemoto and W. J. Munro, Phys. Rev. Lett 93, 250502 (2004). 43. W. J. Munro, Kae Nemoto, T. P. Spiller, S. D. Barrett, P. Kok and R. G. Beausoleil, J. Opt. B: Quantum Semiclass. Opt. 7, S135 (2005). 44. W. J. Munro, K. Nemoto and T. P. Spiller, New J. Phys. 7, 137 (2005). 45. M. G. A. Paris, M. Plenio, D. Jonathan, S. Bose and G. M. D'Ariano, Phys Lett A 273, 153 (2000). 46. G. M. D'Ariano, L. Maccone, M. G. A. Paris and M. F. Sacchi, Phys. Rev. A 61. 053817 (2000). 47. T. B. Pittman, B. C. Jacobs and J. D. Franson, Phys. Rev. Lett. 88, 257902 (2002). 48. J. L. O'Brien, G. J. Pryde, A. G. White, T. C. Ralph and D. Branning, Nature 426, 264 (2003). 49. T. B. Pittman, M. J. Fitch, B. C. Jacobs and J. D. Franson, Phys. Rev. A 68, 032316 (2003). 50. K. Sanaka, T. Jennewein, Jian-Wei Pan, K. Resch and A. Zeilinger, Phys. Rev. Lett. 92, 017902 (2004). 51. S. Gasparoni, J. Pan, P. Walther, T. Rudolph and A. Zeilinger, Phys. Rev. Lett. 93, 020504 (2004). 52. Zhi Zhao, An-Ning Zhang, Yu-Ao Chen, Han Zhang, Jiangfeng Du, Tao Yang and Jian-Wei Pan, Phys. Rev. Lett. 94, 030501 (2005). 53. P. Walther and A. Zeilinger, Phys. Rev. A 72 010302(R) (2005). 54. D. Bouwmeester, Jian-Wei Pan, K. Mattle, M. Eibl, H. Weinfurter and A. Zeilinger, Nature 390, 575 (1997). 55. Jian-Wei Pan, S. Gasparoni, M. Aspelmeyer, T. Jennewein and A. Zeilinger, Nature 421,721 (2003). 56. P. Walther, K. J.Resch, T. Rudolph, E. Schenck, H. Weinfurter, V. Vedral, M. Aspelmeyer and A. Zeilinger, Nature 434, 169 (2005). 57. An-Ning Zhang, Chao-Yang Lu, Xiao-Qi Zhou, Yu-Ao Chen, Zhi Zhao, Tao Yang and Jian-Wei Pan, Experimental Construction of Optical Multi-qubit Cluster States From Bell States, quant-ph/0501036. 58. T. B. Pittman, B. C Jacobs and J. D. Franson, Demonstration of Quantum Error Correction using Linear Optics, quant-ph/0502042. 59. J. L. O'Brien, G. J. Pryde, A. G. White and T. C. Ralph, Phys. Rev. A 71, 060303(R) (2005). 60. A. Furusawa, J. L. Srensen, S. L. Braunstein, C. A Fuchs, H. J. Kimble and E. S Polzik, Science 282, 706 (1998). 61. H. Yonezawa, T. Aoki and A. Furusawa, Nature 431, 430 (2004). 62. A. Dolinska, B. C. Buchler, P. K. Lam, T. C. Ralph and W. P. Bowen, Phys. Rev. A 68, 052308 (2003). 63. N. Takei, H. Yonezawa, T. Aoki and A. Furusawa, Phys. Rev. Lett. 94, 220502 (2005). 64. S. E. Harris and L. V. Hau, Phys. Rev. Lett. 82, 4611 (1999); H. Altug and J. Vucdkovic, Appl. Phys. Lett. 86, 111102 2005. 65. C. Santori, D. Fattal, J. Vuckovic, G.S. Solomon and Y. Yamamoto, Na-
234
ture 419, 594 (2002); E.Waks, K. Inoue, W. D. Oliver, E.Diamanti and Y.Yamamoto, IEEE Journal of Selected Topics in Quantum Electronics 9, 1502 (2003). 66. E. S. Polzik, J. Carry and H. J. Kimble, Phys. Rev. Lett. 68, 3020 (1992).
The contributors of this volume are working at the forefront of various realizations of quantum computers. They survey the recent developments in each realization, in the context of the DiVincenzo criteria, including nuclear magnetic resonance, Josephson junctions, quantum dots, and trapped ions. There are also some theoretical contributions which have relevance in the physical realizations of a quantum computer. This book fills the gap between elementary introductions to the subject and highly specialized research papers to allow beginning graduate students to understand the cutting-edge of research in the shortest possible time.
PHYSICAL REALIZATIONS OF
feQ
QUANTUM COMPUTING Are the DiVincenzo Criteria Fulfilled in 2004? (With CD-Rom)
5928 he ISBN 981-256-473-X
I PUBLISHING
9 "789812 564733"
www.worldscientific.com